5th Annual Symposium on Information Assurance (ASIA ’10 · 2010. 6. 24. · Proceedings of the...
Transcript of 5th Annual Symposium on Information Assurance (ASIA ’10 · 2010. 6. 24. · Proceedings of the...
Academic Track of 13th Annual NYS Cyber Security Conference
Empire State Plaza Albany, NY, USA
June 16-17, 2010
5th Annual Symposium on
Information Assurance (ASIA ’10)
Symposium Chair:
Sanjay Goel Information Technology Management, School of Business University at Albany, State University of New York
conference proceedings
Proceedings of the 5th Annual Symposium on Information Assurance
Academic track of the 13th
Annual 2009 NYS Cyber Security Conference
June 16-17, 2010, New York, USA.
This volume is published as a collective work. Rights to individual papers remain with the author or the author’s
employer. Permission is granted for the noncommercial reproduction of the complete work for educational research
purposes.
Symposium Chairs
Sanjay Goel, Chair Director of Research, NYS Center for Information Forensics and Assurance (CIFA)
Associate Professor, Information Technology Management, School of Business, University at Albany, SUNY
Laura Iwan, Co-Chair State ISO, NYS Office of Cyber Security and Critical Infrastructure Coordination (CSCIC)
Program Committee
Alvaro Ortigosa, Autonoma Universidad de Madrid
Anil B. Somayaji, Carleton University, Canada
Arun Lakhotia, University of Louisiana at Lafayette
Billy Rios, Google, Inc.
Daniel O. Rice, Technology Solutions Experts, Inc.
Dipankar Dasgupta, University of Memphis
George Berg, University at Albany, SUNY
Gurpreet Dhillon, Virginia Commonwealth University
Hemantha Herath, Brock University
Hong C. Li, Intel Corporation
Ibrahim Baggali, Zayed University, Abu Dhabi, UAE
Jeffrey Carr, GreyLogic
John Crain, ICANN
Kenneth Geers, CCD-COE, Talinn, Estonia
Martin Loeb, University of Maryland
Michael Lavine, Johns Hopkins University
Michael Sobolewski, Texas Tech University
Melissa Dark, Purdue University
M.P. Gupta, Indian Institute of Technology, Delhi
Nasir Memon, Polytechnic University of NYU
Pavel Pilugin, Moscow State University, Moscow
Raghu T. Santanam, Arizona State University
Rahul Singh, University of North Carolina, Greensboro
Raj Sharman, University at Buffalo, SUNY
Robert Bangert-Drowns, University at Albany, SUNY
Ronald Dodge, USMA West Point
R. Sekar, Stony Brook University, SUNY
Saeed Abu-Nimeh, Websense, Inc.
Shambhu J. Upadhyaya, University at Buffalo, SUNY
Shiu-Kai Chin, Syracuse University
Siwei Lyu, University at Albany, SUNY
S. S. Ravi, University at Albany, SUNY
Stelios Sidiroglou-Douskos, MIT
Stephen F. Bush, GE Global Research Center
External Reviewers
Manoj Jha, Center of Advanced Transportation and Infrastructure Engineering Research, Morgan State University
Carlo Kopp, Computer Science, Monash University, Melbourne, Australia
Rain Ottis, Cooperative Cyber Defence Center of Excellence, Estonia
Sankardas Roy, Game Theory in Cyber Security Group, University of Memphis
Submissions Chair Damira Pon, University at Albany, SUNY
SYMPOSIUM DINNER SPONSOR
Note of Thanks We would like to express our appreciation to all of the
sponsors which supported the symposium.
CONFERENCE KILOBYTE SPONSOR
Academic Track of 13th Annual NYS Cyber Security Conference
Empire State Plaza Albany, NY, USA
June 16-17, 2010
5th Annual Symposium on
Information Assurance (ASIA ’10)
Symposium Chair:
Sanjay Goel Information Technology Management, School of Business University at Albany, State University of New York
conference proceedings
MESSAGE FROM SYMPOSIUM CHAIRS
Welcome to the 5th
Annual Symposium on Information Assurance (ASIA’10)! This symposium
complements the NYS Cyber Security Conference as its academic track with a goal of increasing
interaction among practitioners and researchers to foster infusion of academic research into
practice. For the last several years, this symposium has been a great success with excellent
papers and participation from academia, industry, and government and highly attended sessions.
This year, we again have an excellent set of papers, invited talks, and keynote addresses.
The keynote speakers this year are Dr. Larry Ponemon and Billy Rios. Dr. Ponemon is the
Chairman and Founder of the Ponemon Institute, a research “think tank” dedicated to advancing
privacy and data protection practices. Billy Rios is currently working as a Security Engineer at
Google, Inc. and has previously had positions at Microsoft, Versign, the Department of Defense,
and E&Y. The symposium has papers in multiple areas of security, including DNS security,
information warfare, risk and security management, intellectual property theft, education and
forensics. We have also included the roundtable discussion on forensics training and education to
the program that we held in previous years based on strong participant interest.
We would like to thank the talented program committee that has supported the review process of
the symposium. In most cases, the papers were assigned to at least three reviewers who were
either members of the program committee or experts outside the committee. We ensured that
there was no conflict of interest and that each program committee member was not assigned to
review more than two papers. We also personally read the papers and the reviews and concurred
with the assessment of the reviewers. For this year’s symposium, we have ten refereed papers
and several invited speakers. Our goal is to keep the quality of submissions high as the
symposium matures. The program committee serves a critical role in the success of the
symposium and we are thankful for the participation of each member.
We were fortunate to have extremely dedicated partners in the NYS Office for Cyber Security
and Critical Infrastructure Coordination (CSCIC) and the University at Albany, State University
of New York (UAlbany). Our partners have managed the logistics for the conference, allowing
us to focus on the program management. We would like to thank the University at Albany’s
School of Business and Symantec for providing financial support for the symposium.
We hope that you enjoy the symposium and continue to participate in the future. In each of the
subsequent years, we plan to have different themes in security-related areas. Next year, again we
are planning on holding the symposium. If you would like to propose a track, please let us know.
The call for papers for next year’s symposium will be distributed in the fall and we hope to see
you again in 2011.
Sanjay Goel Director of Research, CIFA
Associate Professor, School of Business
Laura Iwan NYS ISO, Office of Cyber Security and Critical
Infrastructure Coordination (CSCIC)
i
SYMPOSIUM ON INFORMATION ASSURANCE AGENDA
DAY 1: Wednesday, June 16, 2010
REGISTRATION – Base of the Egg (7:00am – 4:00pm)
VISIT THE EXHIBITORS – Base of the Egg (7:00am – 9:00am)
MORNING SESSION & KEYNOTE – Mtg. Room 6 (9:00am – 11:30am)
Welcome Address: William Pelgrin, Director, CSCIC & George Philip, President, UAlbany
Plenary Talk – Trends in Cyber Crime and Cyber Warfare: Sanjay Goel
Hacking Demo: Sanjay Goel & Damira Pon, School of Business and NYS Center for
Information Forensics and Assurance at the University at Albany, SUNY
Introduction: Peter Bloniarz, CCI, University at Albany, SUNY
Keynote: Dr. Larry Ponemon, Chairman and Founder, Ponemon Institute
LUNCH / VISIT EXHIBITORS – Base of the Egg (11:30am – 12:30pm)
SYMPOSIUM SESSION 1: DNS Security (10:45am – 11:45am) Chair: George Berg, University at Albany, SUNY
Invited Talk: Protecting the Internet
John Crain, ICANN
The Futility of DNSSec
Alex Cowperthwaite & Anil Somayaji, Carleton University
SYMPOSIUM SESSION 2: Information Warfare (1:40pm – 2:40pm) Chair: Stephen F. Bush, GE Global Research
Invited Talk: Big Allied and Dangerous – Terrorist Networks
Victor Asal & R. Karl Rethemeyer, University at Albany, SUNY
Flow-Based Information Warfare Model
Sabah S. Al-Fedaghi, Kuwait University
BREAK / VISIT THE EXHIBITORS – Base of the Egg (2:40pm – 3:00pm)
SYMPOSIUM SESSION 3: Security and Risk Analysis (3:00 – 4:00) Chair: Gurpreet Dhillon, Virginia Commonwealth University
A Multi-Staged Approach to Risk Management in IT Software Implementation
Sunjukta Das, Raj Sharman, Manish Gupta, Varun N. Kutty, Yingying Kang
University at Buffalo, SUNY
A Survivability Decision Model for Critical Information System Based on Bayesian
Network
Suhas Lande, Yunjun Zuo, & Malvika Pimple, University of North Dakota
ii
SYMPOSIUM ON INFORMATION ASSURANCE AGENDA, CONT’D.
DAY 2: Thursday, June 17, 2010
REGISTRATION / VISIT EXHIBITORS – Base of Egg (7:30am – 4:00 pm)
ASIA Keynote & Best Paper Award – Mtg. Rm. 6 (9:00am – 10:15am)
Introduction: Sanjay Goel, Symposium Chair
Welcome Address: Donald Siegel, Dean, School of Business, UAlbany
So You Wanna Be a Bot Master?
Billy Rios, Google, Inc.
Best Paper Award Presentation: Laura Iwan, Symposium Co-Chair
SYMPOSIUM SESSION 4: Intellectual Property Theft (10:30 – 11:30am) Chair: Anil Somayaji, Carleton University
A Week in the Life of the Most Popular BitTorrent Swarms Mark Scanlon, Alan Hannaway, & Mohand-Tahar Kechadi, University College Dublin, Centre
for Cybercrime Investigations
Information Assurance in Video Streaming: A Case Study
Sriharsha Banavara, Eric Ismand, & John A. Hamilton, Jr., Auburn University
LUNCH / VISIT THE EXHIBITORS – Base of the Egg (11:30am – 12:30pm)
SYMPOSIUM SESSION 5: Security Management (12:30 – 1:30pm) Chair: Raj Sharman, University at Buffalo, SUNY
Assessing the Impact of Security Culture and the Employee-Organization Relationship in
IS Security Compliance Gwen Greene, Towson University & John D’Arcy, University of Notre Dame
Invited Paper: Employee Emancipation and Protection of Information Yurita Abdul Talib & Gurpreet Dhillon, Virginia Commonwealth University
BREAK / VISIT THE EXHIBITORS – Base of the Egg (1:30pm – 1:50pm)
SYMPOSIUM SESSION 6: Security Education (1:50pm – 2:50pm) Chair: Siwei Lyu, University at Albany, SUNY
Enhancing Security Education with Hands-On Laboratory Exercises
Wenliang Du, Karthick Jayaraman, & Noreen B. Gaubatz, Syracuse University
A Digital Forensics Program to Retrain America’s Veterans
Eric Ismand & John A. Hamilton, Jr., Auburn University
SYMPOSIUM SESSION 7: Roundtable: Forensic Education (3:00pm – 3:50pm)
Panelists: Fabio Auffant, NY State Police; Kevin Williams, Psychology, UAlbany; John A.
Hamilton, Jr., Auburn University
CLOSING REMARKS (3:50pm – 4:00pm) Sanjay Goel, Symposium Chair
iii
TABLE OF CONTENTS
Session 1: DNS Security
Invited Talk: Protecting the Internet ………………………………………….…..............
John Crain, ICANN
The Futility of DNSSec ………………………………………….….....................................
Alex Cowperthwaite & Anil Somayaji, Carleton University
1
2
Session 2: Information Warfare
Invited Talk: Big Allied and Dangerous – Terrorist Networks …………………..…..…
Victor Asal & R. Karl Rethemeyer, University at Albany, SUNY
Flow-Based Information Warfare Model ………………………….…………….……….
Sabah S. Al-Fedaghi, Kuwait University
9
10
Session 3: Security and Risk Analysis
A Multi-Staged Approach to Risk Management in IT Software Implementation……... Sunjukta Das, Raj Sharman, Manish Gupta, Varun N. Kutty, & Yingying Kang
University at Buffalo, SUNY
19
A Survivability Decision Model for Critical Information System Based on Bayesian
Network ………………………………………………….………………………………….
Suhas Lande, Yunjun Zuo, & Malvika Pimple, University of North Dakota
23
Symposium Keynote Address
So You Wanna Be a Bot Master? ……………………………………………………….…
Billy Rios, Google, Inc.
31
Session 4: Intellectual Property Theft A Week in the Life of the Most Popular BitTorrent Swarms ….……….……………......
Mark Scanlon, Alan Hannaway, & Mohand-Tahar Kechadi, University College Dublin, Centre
for Cybercrime Investigations
Information Assurance in Video Streaming: A Case Study ……………………………...
Sriharsha Banavara, Eric Ismand, & John A. Hamilton, Jr., Auburn University
32
37
Session 5: Security Management
Assessing the Impact of Security Culture and the Employee-Organization Relationship in
IS Security Compliance …………………………………………………..………………
Gwen Greene, Towson University & John D’Arcy, University of Notre Dame
Invited Paper: Employee Emancipation and Protection of Information ………………..
Yurita Abdul Talib & Gurpreet Dhillon, Virginia Commonwealth University
42
50
SESSION 6: Security Education
Enhancing Security Education with Hands-On Laboratory Exercises ………………….
Wenliang Du, Karthick Jayaraman, & Noreen B. Gaubatz, Syracuse University
A Digital Forensics Program to Retrain America’s Veterans …………………...……….
Eric Ismand & John A. Hamilton, Jr., Auburn University
56
62
SESSION 7: Roundtable: Forensic Education…………………………………………………. Panelist: Fabio Auffant, NY State Police; Kevin Williams, UAlbany; John A. Hamilton, Jr.,
Auburn University
67
Author Biographies…………………………………………………………………………. 68
Index of Authors ……………………………………………………………………...……... 73
Invited Talk: The Changing Landscape of the DNS
John Crain, Chief Technical Officer ICANN (Internet Corporation for Assigned Names and Numbers)
HE DNS is one of the things that people rarely think about, or even know about, when they use the
Internet. John will give you an update of some of the things you may notice and some of the things you
won’t. There are some things that go on behind the scenes of the management of the DNS that hardly ever
get noticed that will be revealed in this talk. His update will also touch upon what may be two of the
largest changes since the DNS was introduced. 1) DNSsec, which has enabled a new layer of security, and
2) IDN, which allows navigation of the network in many scripts. These advances will at some point affect
everyone who uses the Internet.
T
Abstract—The lack of data authentication and integrity
guarantees in the Domain Name System (DNS) facilitates a wide
variety of malicious activity on the Internet today. DNSSec, a set
of cryptographic extensions to DNS, has been proposed to address
these threats. While DNSSec does provide certain security
guarantees, here we argue that it does not provide what users
really need, namely end-to-end authentication and integrity. Even
worse, DNSSec makes DNS much less efficient and harder to
administer, thus significantly compromising DNS’s availability—
arguably its most important characteristic. In this paper we
explain the structure of DNS, examine the threats against it,
present the details of DNSSec, and analyze the benefits of DNSSec
relative to its costs. This cost-benefit analysis clearly shows that
DNSSec deployment is a futile effort, one that provides little long-
term benefit yet has distinct, perhaps very significant costs.
I. INTRODUCTION
HE Domain Name System (DNS) has become an vital part
of the modern Internet. DNS translates human readable
host names such as www.carleton.ca to machine routable IP
addresses such as 134.117.1.32. These operations are
performed automatically as part of nearly every transaction
across the Internet today.
DNS was designed in the early 1980’s to be fast, efficient,
and highly available in the face of network disruptions [1][2].
These were all essential design goals given the limited and
unreliable nature of the computational and networking
resources at their disposal. To achieve these goals DNS used a
distributed hierarchical database, a simple query response
protocol, and aggressive caching with permissibly stale data.
With the rapid and global growth of the Internet, the DNS
system has scaled very well and maintained its core attributes
of efficiency and high availability; unfortunately, it has also
maintained its lack of data authenticity and integrity. These
omissions have become increasingly significant as malicious
activity has increased on the Internet. A falsified DNS reply
can redirect an unsuspecting user to an attacker-controlled
host, making them vulnerable to phishing, malware
distribution or numerous other malicious activities.
DNS Security Extensions (DNSSec) addresses DNS’s lack
of data authentication and integrity. DNSSec provides these
This work was supported by Canada’s Natural Sciences and Engineering
Research Council (NSERC) through the Internet-worked Systems Security
Network (ISSNet) and Discovery Grant program.
guarantees by using public key cryptography to sign each DNS
record. DNSSec has been carefully designed to be as efficient
and scalable as possible; as we will explain, however, the
bandwidth, computational, and administrative overhead of
DNSSec do potentially compromise DNS’s essential
availability properties. More significant, however, is that
DNSSec solves the wrong problem: it secures hostname to IP
address mappings when what is really needed is better end-to-
end security guarantees. Thus we believe that the deployment
of DNSSec is a futile gesture, one that will lead to minimal
long term security benefits while resulting in significant
security and economic costs.
We proceed to make this argument in the rest of this paper
as follows. First, we describe the Domain Name System in
more detail in Section II. Next, we discuss cache poisoning
attacks, the most serious current threat to DNS, in Section III.
Section IV presents DNSSec and explains how DNSSec can
block cache poisoning attacks. We then explore how DNSSec
reduces the availability of DNS in Section V; Section VI
discusses how attackers could circumvent the protections
provided by DNSSec. Section VII presents potential
alternatives to DNSSec adoption while Section VIII discusses
more general issues relating to DNSSec and end-to-end
security. Section IX concludes.
II. THE DOMAIN NAME SYSTEM
The Domain Name System is a distributed database for
storing information on domain names, the primary namespace
for hosts on the Internet. While the primary purpose of DNS is
to map domain names to IP addresses, it actually supports
several kinds of records. The primary record types are as
follows:
1. A records specify an IPv4 address for a host.
2. AAAA records specify an IPv6 address for a host.
3. NS records point to the authoritative name servers
for a zone (domain).
4. CNAME records allow the specification of host
aliases, e.g. make www.example.com refer to
berners-lee.example.com.
5. MX records identify mail servers that can receive
email for a domain.
The DNS system is composed of two types of systems,
name servers and resolvers. Name servers return information
The Futility of DNSSec
Alex Cowperthwaite and Anil Somayaji School of Computer Science, Carleton University
Ottawa, ON Canada K1S 5B6
{acowpert, soma}@ccsl.carleton.ca
T
about domains or zones when queried by DNS resolvers. DNS
resolvers send queries to one or more name servers in order to
service DNS requests from applications.
When a name server is configured as authoritative for its
zone, its resource records are always interpreted as correct. To
improve availability and reduce upstream server load,
resolvers may also cache resource records until they have
expired; while some domains specify time-to-live values on the
order of minutes, it is more common for domains to permit
their records to be cached for days. Cached records are not
considered to be authoritative, i.e., they may contain incorrect
or stale information; most consumers of DNS information,
however, do not distinguish between authoritative and non-
authoritative results.
Applications normally send queries to simple stub resolvers
implemented as library functions. Stub resolvers forward
requests to standalone recursive resolvers that perform the
multiple name server queries normally required to answer any
request. Recursive resolvers work best when most responses
can be answered using cached queries (i.e., they see multiple
requests for the same information). ISPs normally operate
DNS servers with recursive, caching resolvers for their
customers; alternately, users can configure their systems to
connect to other open DNS servers1.
DNS name servers are structured as a hierarchical tree
where the tree levels correspond to the dot-separated
components of a domain name. A complete lookup follows a
chain starting from the DNS root servers to a top level domain
(TLD) down to any number of domains and sub-domains
below. We review the steps in a typical DNS lookup below.
1. An application (e.g., a web browser) invokes a local
stub resolver to look up the IP address of
www.example.com.
2. The stub resolver sends the request to a local DNS
server (a recursive, caching DNS resolver).
3. The DNS server will check its cache for an entry for
www.example.com. If it is does not have an entry it
will request a lookup from a randomly chosen root
server. The IP addresses of standard root servers
are built in to DNS servers.
4. The root server will reply with the NS record
describing the name of the authoritative name
server for the Top Level Domain (TLD) .com. To
reduce traffic, it will also append ―the glue‖: the A
record (IPv4 address) for the .com TLD name
server.
5. The recursive DNS resolver will request a lookup
for www.example.com from the .com authoritative
name server (as specified in the returned glue A
record).
1 Traditionally, recursive resolvers and name servers were both implemented
as part of the same application (e.g., BIND 9 and earlier [31]; Newer DNS
server designs, however, divide these functions into separate programs (e.g.
tinydns and dnscache of djbns [4]).
6. The .com name server will reply with the
authoritative name server record for example.com.
It will also append the A record for the
example.com authoritative name server.
7. The recursive DNS resolver will then ask the
authoritative name server for example.com for the
lookup of www.example.com
8. The name server for example.com will reply with an
authoritative answer for www.example.com, likely
only an A record.
9. Optionally, the recursive DNS resolver will add or
update this entry in its cache.
10. The recursive DNS server will then return a non-
authoritative answer (an A record) to the
requesting application’s stub resolver.
Each DNS query and response is normally sent in a single
UDP packet. The query and response use the same structure;
the query is sent with the response field blank. A 16 bit query
ID field used to distinguish outstanding requests [5][6].
Note how multiple characteristics of DNS make it both
efficient and highly available. Queries are simple and small
(there are few record types and the amount of information in
each is small), minimizing bandwidth and storage overhead.
Full and partial results are easily cached (i.e., servers cache
results from every level of the hierarchy). And, DNS’s
database is distributed in a simple, scalable way that is
(relatively) easily debugged and tuned—i.e., if you don’t get
an appropriate response, you can quickly identify what server
gave you incorrect information.
III. DNS CACHE POISONING
While DNS has many positive characteristics, note that
DNS has no integrity, authenticity, or confidentiality
guarantees. Confidentiality can be added by encrypting queries
(e.g., via IPsec); integrity and authenticity cannot be so easily
added, however, because of the distributed nature of DNS.
DNS responses can be modified or completely forged almost
trivially by an intruder-in-the-middle attack. Indeed, because
DNS normally uses UDP, attackers do not even need to hijack
TCP connections. The weaknesses of DNS are actually even
worse, however, because the existing DNS infrastructure can
be used to deliver malicious responses.
Specifically, malicious responses can originate from
legitimate, otherwise uncompromised servers through the use
of cache poisoning attacks [7]. With standard DNS cache
poisoning attacks, a victim can be redirected to a malicious
host through the injection of a forged A record. The key idea
behind such attacks is that a caching resolver has no way to
authenticate received data; it will accept any query response
that ―looks right.‖ An attacker even has many chances to
generate authentic-looking responses as unexpected responses
will simply be discarded.
While such attacks are problematic enough, in the summer
of 2008 Dan Kaminsky described a direct extension to known
DNS cache poisoning techniques that made them much more
dangerous and reliable. Kaminsky’s primary observation was
that instead of spoofing the final A record of the destination,
an attacker could spoof any one of the query responses in the
lookup process. By forging an NS record, an attacker could
take over an entire domain, up to and including a TLD such as
.com.
Kaminsky also improved the reliability of cache poisoning
attacks. Previously, if an entry was cached attackers would
have to wait for the cached entry to expire before they could
attempt to replace it; thus, it was relatively hard to poison the
cache of popular domains. Kaminsky, however, noted that by
querying an obscure sub-domain not already cached (i.e.
reallyobscuresubdomain.example.com) an attacker could force
a lookup. The lookup would go directly to the authoritative
nameserver for the domain, assuming it was cached. This
creates an opportunity for an attack to forge a reply with no
answer for the obscure sub-domain but include an additional
update for the A record of the authoritative nameserver. The
opportunistic caching resolver will update their cache with the
false record. This method allows an attacker to repeat the
attack many times giving them a greater chance of returning an
answer that will be accepted, successfully poisoning the cache
[8].
The easiest-to-implement defense against these cache
poisoning attacks is to make responses harder to predict.
Clearly the 16-bit query ID field does not have enough entropy
to prevent brute force attacks. This entropy can be
supplemented by randomizing the outgoing request port. If
every available port could be used, that would give us 32 bits
(as UDP ports are 16 bit quantities); while we cannot reach
this upper bound in practice, we can get close enough to make
cache poisoning attacks much harder. Even with such
improvements, however, it is still possible to mount DNS
cache poisoning attacks. To prevent forged requests, ideally
we need the kind of authentication guarantees that are best
provided by cryptographic mechanisms. DNSSec was
designed to provide precisely these guarantees.
IV. DNSSEC
DNSSec is a set of extensions to DNS that are designed to
authenticate and protect the integrity of DNS responses
through the use of public key-based digital signatures. The
design of DNSSec started in the mid-1990s, motivated by the
growing recognition of the dangers of DNS cache poisoning
[7]. The first version of DNSSec was officially published as
RFC 2065 in January 1997 [9]; an improved version followed
in RFC 2535 (March 1999) [10]; after trial deployment
experience, the current version of DNSSec was finalized in
RFC 4033-4035 in March 2005 [11][12][13].
DNSSec allows domain owners to digitally sign resource
records. One key design decision in DNSSec is that name
servers do not implement any cryptography; signatures are
generated offline, while signatures are checked by resolvers.
While this design choice minimizes the computational
overhead of DNSSec, it also greatly complicates the process of
authenticating DNS responses because the results returned by
every authoritative name server in a query chain must be
individually verified. Thus, for resolving www.example.com,
the responses from the root, .com, and example.com name
servers must each be authenticated independently. The
information necessary to authenticate responses is stored in a
set of new DNS record types:
RRSIG records contain digital signatures for other
resource records. Each regular DNS resource
record should have its own RRSIG record (except
for the DNSKEY record).
DNSKEY records contain public keys for verifying
RRSIG records. Note that a zone may have
multiple DNSKEY records.
DS, or designated signer records, store
cryptographic hashes of the public keys of a zone’s
children. DS records are what allow .com to
specify what key should sign example.com’s
records, thus enabling trust chains. DS records
should be signed with a corresponding RRSIG
record.
NSEC and NSEC3 records provide authenticated
denial of existence. NSEC provided this
information by specifying two hostnames between
which no other hostnames exist. NSEC allows
attackers to enumerate the hosts in a domain;
NSEC3 [14] addresses this problem by specifying
the ranges in terms of hashes of domain names
rather than the domain names themselves.
To authenticate a resource record, a resolver first checks
that the RRSIG signature on the requested record was
generated by the zone’s key, as specified by its DNSKEY
record. To make sure the DNSKEY record has the right value,
the corresponding DS record of the zone’s parent is checked to
see if it has the hash of the DNSKEY key. Then, to verify the
DS record is genuine, its RRSIG has to be checked against the
parent zone’s DNSKEY public key. This process repeats until
we get to the root zone; here, we assume the resolver already
knows the authentic public keys belonging to the root name
servers.
Note that DNSSec clearly defeats cache poisoning attacks.
Responses can only be forged if the underlying cryptography is
broken. Because resolvers will only forward or cache authentic
responses, the resolver’s cache would remain uncorrupted.
Unfortunately this benefit comes along with significant costs.
V. DNSSEC AVAILABILITY ISSUES
While DNSSec does provide strong integrity and
authenticity guarantees for DNS responses, it also requires
significantly more computer resources and requires a
significant investment in key management. Both of these
factors reduce DNS’s availability and so counterbalance the
potential security gains from DNSSec.
A. Resource Usage
DNSSec significantly increases the computational and space
footprint of DNS. DNSSec requires additional computation for
generating and verifying records. These costs are either offline
or are borne by requesting clients and/or their local resolvers;
as such, cryptographic overhead shouldn’t be a problem for
most DNSSec deployments.
The most obvious source of increased space usage with
DNSSec is the use of additional records: every regular record
requires a RRSIG signature, along with at least one
nameserver-specific DNSKEY and verifying DS (from the
parent) records. While the space impact of this increase isn’t
too large given the relatively small size of DNS zone files,
these additional records must also be communicated,
increasing DNS’s bandwidth requirements significantly.
Given today’s computers and networks, this increased
resource consumption should be very manageable under
normal circumstances. Root and TLD DNS servers, however,
have been and will continue to be targets of Distributed
Denial-of-Service (DDoS) attacks [15][16].
The increased bandwidth requirements of DNSSec will
require significant hardware, especially bandwidth, upgrades
to maintain the same level of DDoS resistance. Any outage or
latency at the root or TLD servers will be felt throughout the
internet. The increased bandwidth consumption is particularly
worrisome because of its asymmetric nature: small DNS
requests can result in large DNSSec replies. This characteristic
makes DNSSec-enabled systems more attractive for DNS
DDoS amplification attacks. Such attacks happen when a
DDoS attack is passed through DNS servers to amplify the
volume of data being sent to a target. Amplification attacks are
feasible because it is so easy to forge the source address of
UDP packets; thus, any open resolver can be used to send
packets to almost any host on the Internet [17][18].
To evaluate the possibility of the amplification attack, we
recorded DNS lookups using Wireshark and the DNS lookup
utility dig (See Table 1). We looked up the host
www.dnssec.se using the L root server and the open .se
DNSSec test server. We chose these servers because they offer
some of the strongest DNSSec support at the time of this
writing. We found that many servers do not fully implement
DNSSec as RFC 4033–4035 specify, so we report both the
actual values observed on the wire and the expected values
based on the RFCs. Specifically, we manually reconstructed
the full chain of DNS replies that is supposed to be returned
when the checking disabled (CD) flag is set. From our data it
is clear that DNSSec greatly increases the opportunity for an
amplificaion attack.
DNS amplification attacks can be mitigated in multiple
ways. DNS resolvers can be shielded from external requests
through firewall rules or through IP address-based request
filters on DNS servers themselves. Unicast Reverse Path
Forwarding (URPF), defined in RFC 3074 [19], limits the
scope of IP address spoofing. Such defenses, however, have
only been partially deployed; further, even where deployed
limited forms of IP address spoofing are still possible. With
DNSSec deployment attackers will need far fewer DNS
amplifiers to perform potent denial of service attacks, making
DNS amplification attacks that much more feasible.
While neither the increased DDoS susceptibility and the
increased potential for DNS DDoS amplification attacks
should, by themselves, stand in the way of DNSSec
deployment, they are both clear drawbacks to a transition to
DNSSec.
B. Key Management
We think a more significant DNSSec disadvantage will be
key management in practice. DNSSec has a number of
provisions designed to make key management easier. It allows
a zone to have as many signing keys (DNSKEY records) as
they wish, and those keys can be divided into separate
functions, e.g., Zone Signing Keys (ZSK) and Key Signing
Keys (KSK). The fundamental fact is, however, is that DNS
administrators will have to become public-key infrastructure
(PKI) administrators as well. They will have to secure private
signing keys, decide who has access to them, and generate
signatures for new and old information as records change and
as keys expire. DNS administrators are already busy people;
certificate management is a new, non-trivial responsibility for
which many (arguably most) DNS administrators do not have
the background, training, or experience. The most obvious
potential issue with such inexperienced DNSSec
administrators is that they will store and use keys insecurely,
making themselves vulnerable to attack. While such direct
attacks may be a concern, we think such issues will be minor
compared to two others: DNSSec-caused denial of service and
false alarms.
Ideally, DNSSec would be used such that any signature
verification error would cause a DNS lookup to fail. If
DNSSec is deployed this way, however, any key management
or signature generation error would result in a denial of
service. Sign using the wrong key—site goes down. Sign with
an expired key—site goes down. Forget to update signatures
when a key expires—site goes down as well. Anecdotally we
know users occasionally experience certificate verification
errors with SSL [20]; when every domain on the Internet must
manage keys, and when every hostname must be appropriately
signed, user-visible key management errors will inevitably go
up dramatically.
We hypothesize that these errors will be so common that it
will not be feasible for DNSSec to cause DNS queries to fail;
TABLE I
AMPLIFICATION RELATIVE TO ONE 70-BYE
QUERY PACKET FOR WWW.DNSSEC.SE
No. of
Packets
Total Size
(bytes) Total Size
(bytes)
DNS Reply 1 208 2.9
DNSSec Reply (Actual) 1 1215 17.3
DNSSec Reply (Expected) 1 2257 32.2
DNSSec Reply Chain (Actual) 5 3045 43.5
DNSSec Reply Chain (Expected) 5 11865 169.5
instead, DNSSec authentication errors will be used to generate
warnings that users or administrators can choose to follow or
ignore—just as with SSL certificates. And, as with SSL
certificates, most users will choose to ignore these warnings
[21].
In other words, key management with DNSSec will be
problematic enough that either users will have to tolerate a
significant loss in system availability due to non-malicious
authentication failures, or they will become accustomed to
ignoring DNSSec warnings, eliminating any security benefit
DNSSec provides.
VI. CIRCUMVENTING DNS SECURITY
When properly deployed, DNSSec does guarantee integrity
and authenticity of DNS data. Even if we assume that DNSSec
is used optimally, however, that does not mean that we get
what we really want, namely secure communications.
Attackers have two basic strategies for circumventing
DNSSec: they can attack from below or from above. Attackers
can get ―below‖ DNSSec by targeting other aspects of network
communication. DNS, at best, tells a computer the IP address
of another computer. An attacker who mounts an intruder-in-
the-middle attack (using, say, ARP spoofing on a local
network [22] or BGP route hijacking [23][24]) can potentially
take over any IP address they desire. Thus, DNS data can be
completely secure yet attackers can still make victim machines
connect to malicious hosts. Targeting the actual IP
communications with the host instead of the DNS lookup, an
attacker easily circumvents any security gained through
DNSSec.
Of course, all this assumes that the DNS hostname we are
attempting to resolve is the right hostname. Phishing attacks
have demonstrated that users are easily fooled into visiting
malicious sites via links containing fraudulent hostnames [25].
Users do not understand the significance of hostname
components and their ordering. Indeed, why shouldn’t a
regular user visit a URL with www-mybank.secure.com when
their bank is really located at www.mybank.com? A proper
explanation requires a basic understanding of DNS—
something that we cannot expect regular users to know about.
DNSSec is a (potentially) strong defense surrounded by
insecurity both above and below. While DNSSec deployment
might reduce the incidence of DNS-specific attacks, we cannot
see how it would have a significant impact on the number or
severity of security incidents on the Internet. It is simply too
easy for an attacker to use other means to get systems and
users to connect to malicious hosts.
VII. BEYOND DNSSEC
While there are alternatives to DNSSec such as DNSCurve
[26] that are more efficient and that arguably have better
security properties, to address the above concerns we need to
realize that DNS, by itself, is not the security problem that we
need to address. What is needed are better ways to provide
end-to-end security: when a person uses a computer to interact
with a remote host, we need ways to authenticate that remote
host to defend against intruder-in-the-middle attacks and
attacker attempts to masquerade as that remote host. DNSSec
does not provide end-to-end authentication because IP-level
traffic is not authenticated and because people can easily
confuse malicious domain names with legitimate ones (or,
they’ll just ignore malicious domain names).
Today we have two technologies in wide use for providing
end-to-end authentication. One, SSL/TLS [27] as it is
implemented on the web, authentication occurs using trust
chains grounded in public keys bundled with web browsers.
The other is Secure Shell (SSH) [28], where the public keys of
remote hosts are either cached or supplied by an administrator.
With SSL, a third party (a certificate authority), generally
unknown to a user, vouches that the host being accessed is the
one denoted by its domain name (e.g., Verisign says that we
are connecting to www.bankofamerica.com when browsers
visit that domain), and with the new extended validation
certificates, that third party says what organization controls
that server (e.g., Verisign says that www.bankofamerica.com
belongs to Bank of America Corporation). Alternately, with
SSH on our first connection we cannot authenticate the host;
however, on subsequent connections SSH authenticates the
remote host by correlating the domain name and server-
supplied public key with the key information that was saved
previously. So long as you connect to the same hosts you have
in the past, you are protected from attackers masquerading or
intercepting traffic at the network level.
SSH gives us easy-to-use end-to-end authentication (and
more) so long as we connect to the same remote hosts; it
provides no authentication for connections to new hosts. SSL
gives us remote host authentication so long as we trust the
certificate authority keys included with our browsers, we
verify the domain name we are visiting is, indeed, the correct
one, and we cancel connections to hosts that generate
certificate errors. Of course, all three of these assumptions are
not actually true: sometimes certificate authorities are
organizations that many people distrust [29][30] and users
regularly ignore malicious URLs and certificate errors [25]. As
we discussed earlier, however, these issues also apply to
DNSSec or, indeed, any scheme that relies on the DNS
namespace and arbitrary third-party signers of keys.
Both SSH and SSL are built on top of DNS and they both
provide the authentication DNS lacks. Even better, they
provide host authentication, not host IP address authentication,
although with some issues. Thus, it is reasonable to argue that
SSH and SSL together provide remote host authentication for
all users who want or need such authentication, and they do so
in a way that also provides end-to-end security, thus
obliviating the need for DNSSec. Yet neither address the
bigger problem of bridging the gap between hostnames and the
remote host (and organization) a user actually intends to
access.
VIII. DISCUSSION
The Domain Name System is efficient, fault tolerant, and
fundamentally insecure—much as is the case with TCP/IP.
Both are foundational technologies for the Internet. Since the
mid-1990’s, many researchers have worked on developing
secure variants of both protocols, namely DNSSec and IPsec;
neither of these more secure proposals have been widely
deployed. Note that while DNSSec by itself can be
circumvented by IP-level attacks, IPsec and DNSSec together
provide a very secure foundation. So, why haven’t they been
widely deployed?
Part of the reason is simple laziness (in the most reasonable
sense): both of these transitions would involve significant
ongoing cost because of the burden of key management, and
without widespread deployment their security benefits are
minimal. The more fundamental reason, though, is that
complete layer-level changes are simply infeasible today on
the Internet. They may happen over time, but only if
circumstances are just right. The normal case is to expect
fragmentation—some will adopt the new technology while
others won’t. IPv6 and software patches are in the same boat
as DNSSec and IPsec— some will deploy but all will not. In
this environment, the more a solution requires universal or
near-universal acceptance to provide benefits, the more likely
it will never provide those benefits.
While DNS cannot provide end-to-end security without a
cryptographic extension such as DNSSec, we could make DNS
immune to cache poisoning attacks, the main source of DNS
insecurity, with a variety of relatively minor changes (e.g., by
increasing the size of the query ID field from 16 to 64 bits)—
that is, assuming that the existing best practice of randomizing
source ports and query IDs is not sufficient protection. To get
end-to-end security, we simply need to encourage more sites to
adopt SSL or, as appropriate, SSH. SSL is already ubiquitous
and SSH provides benefits even in limited deployments.
Some advocates of DNSSec are arguing that it should be
adopted not because of the security guarantees it provides, per
se, but because of the new, more secure applications that could
be built on top of it [31][32]. But this argument assumes
widespread deployment, something that has not happened and
may never happen in a way that allows other applications to be
built on top of it. Rather than waiting for a technology that has
been in development for fifteen years to suddenly take off,
why not use existing technologies to achieve the same ends?
But pushing for better end-to-end security at the DNS level
(using DNSSec, SSL, or SSH) misses the larger security issues
of authenticating destinations for network communications,
particularly web-based communications initiated by users. The
fundamental issue here is that the hostnames namespace is not
one that regular people understand. People understand brands
and, to a lesser extent, company names. Given the numerous
and still growing number of top-level domains, there exists no
clean mapping between brands, companies, and hostnames.
Many Internet users have discovered a very simple strategy for
dealing with this confusion: they do not access sites via
domain names and URLs; instead, they use search engine
queries. Such use of search engines can sometimes lead to
users accessing the wrong site (this happened with Facebook
recently [33]). Accessing hosts via search engines is
potentially insecure because there is no guarantee that the link
the search engine returns is the one the user actually wants;
however, this method is fast, relatively reliable, and
understandable to end users. If we wish to secure Internet
communications, we must take into account how computer
users actually behave, not how we would like them to behave.
What we need are ways for users to specify remote hosts
precisely and securely, using a namespace that does not
require a computer science degree to understand and properly
use. We then need a way to handle authentication failures in a
way that people are more likely to follow, not ignore them.
These are hard questions, ones that need further research. The
arguments we have made against DNS and DNSSec have been
made separately by many others; our contribution here is to
bring them together and to put them in perspective versus the
real security problem of end-to-end security. For
administrators, we hope this work will help them better
evaluate security technologies such as DNSSec. Also, we hope
that once researchers see the limited benefits and the
significant drawbacks of DNSSec that they will move on to
developing incrementally deployable solutions that address the
true end-to- end security needs of users on today’s Internet.
IX. CONCLUSION
DNSSec is designed to improve Internet security by
providing authentication and integrity protection for DNS.
While DNSSec would potentially protect against DNS cache
poisoning, it does so at the cost of increased susceptibility to
DDoS attacks, increased utility for DDoS amplification
attacks, and significant human administrator overhead for key
management. The key management issue is particularly
important because poor key management (which we expect to
be the rule, not the exception) will result in denials of service
or, more likely, in users being trained to ignore DNSSec-based
warnings.
These significant tradeoffs take place against a backdrop of
other, simpler methods for defending against cache poisoning
attacks such as source-port randomization and DNSSec’s
inability to provide end-to-end authentication. While DNSSec
could be extended to provide such protection, we already have
solutions that provide essentially the same guarantees: SSL
and SSH.
The true futility of DNSSec, however, arises from the fact
that it addresses the wrong problem. Users do not understand
the semantics of DNS well enough to distinguish between
malicious and legitimate hosts. While extended validation SSL
certificates are a small step in the right direction, interface
issues [34] and user reliance on search engines instead of
URLs [33] make them a partial solution at best.
Rather than adopt DNSSec, we argue that the Internet
community should invest in making DNS more robust to cache
poisoning attacks without relying upon a PKI infrastructure
while SSL and SSH should be more widely adopted.
Meanwhile, research is needed in ways to close the gap
between user perception and remote host authentication.
ACKNOWLEDGMENT
We wish to thank the members of Carleton Computer
Security Laboratory and the anonymous reviewers for their
suggestions.
REFERENCES
[1] P. Mockapetris, ―Domain names: Concepts and facilities,‖ RFC 882,
Internet Engineering Task Force, Nov. 1983. [Online]. Available:
http://www.ietf.org/rfc/rfc882.txt
[2] ——, ―Domain names: Implementation specification,‖ RFC 883,
Internet Engineering Task Force, Nov. 1983. [Online]. Available:
http://www.ietf.org/rfc/rfc883.txt
[3] I. S. Consortium, ―Isc bind.‖ [Online]. Available: https://www.isc.org/
software/bind
[4] D. J. Bernstein, ―dnjdns.‖ [Online]. Available:
http://cr.yp.to/djbdns.html
[5] P. Mockapetris, ―Domain names - concepts and facilities,‖ RFC 1034
(Standard), Internet Engineering Task Force, Nov. 1987, updated by
RFCs 1101, 1183, 1348, 1876, 1982, 2065, 2181, 2308, 2535, 4033,
4034, 4035, 4343, 4035, 4592. [Online]. Available:
http://www.ietf.org/rfc/rfc1034.txt
[6] ——, ―Domain names - implementation and specification,‖ RFC 1035
(Standard), Internet Engineering Task Force, Nov. 1987, updated by
RFCs 1101, 1183, 1348, 1876, 1982, 1995, 1996, 2065, 2136, 2181,
2137, 2308, 2535, 2845, 3425, 3658, 4033, 4034, 4035, 4343. [Online].
Available: http://www.ietf.org/rfc/rfc1035.txt
[7] S. Bellovin, ―Using the domain name system for system break-ins,‖ in
The 5th Usenix UNIX Security Symposium. Usenix, June 1995.
[8] S. Friedl, ―An illustrated guide to the kaminsky dns vulnerability,‖ Aug
2008. [Online]. Available: nurlfhttp://unixwiz.net/techtips/ iguide-
kaminsky-dns-vuln.htmlg
[9] D. Eastlake 3rd and C. Kaufman, ―Domain Name System Security
Extensions,‖ RFC 2065 (Proposed Standard), Internet Engineering Task
Force, Jan. 1997. [Online]. Available:
http://www.ietf.org/rfc/rfc2065.txt
[10] D. Eastlake 3rd, ―Domain Name System Security Extensions,‖ RFC
2535 (Proposed Standard), Internet Engineering Task Force, Mar. 1999.
[Online]. Available: http://www.ietf.org/rfc/rfc2535.txt
[11] R. Arends, R. Austein, M. Larson, D. Massey, and S. Rose, ―DNS
Security Introduction and Requirements,‖ RFC 4033 (Proposed
Standard), Internet Engineering Task Force, Mar. 2005. [Online].
Available: http://www.ietf.org/rfc/rfc4033.txt
[12] ——, ―Resource Records for the DNS Security Extensions,‖ RFC 4034
(Proposed Standard), Internet Engineering Task Force, Mar. 2005,
updated by RFC 4470. [Online]. Available: http://www.ietf.org/rfc/
rfc4034.txt
[13] ——, ―Protocol Modifications for the DNS Security Extensions,‖ RFC
4035 (Proposed Standard), Internet Engineering Task Force, Mar. 2005,
updated by RFC 4470. [Online]. Available: http:
//www.ietf.org/rfc/rfc4035.txt
[14] B. Laurie, G. Sisson, R. Arends, and D. Blacka, ―DNS Security
(DNSSEC) Hashed Authenticated Denial of Existence,‖ RFC 5155
(Proposed Standard), Internet Engineering Task Force, Mar. 2008.
[Online]. Available: http://www.ietf.org/rfc/rfc5155.txt
[15] D. Moore, C. Shannon, D. Brown, G. Voelker, and S. Savage, ―Inferring
Internet denial-of-service activity,‖ ACM Transactions on Computer
Systems (TOCS), vol. 24, no. 2, p. 139, 2006.
[16] N. Brownlee, K. Claffy, and E. Nemeth, ―DNS measurements at a root
server,‖ GLOBECOM-NEW YORK-, vol. 3, pp. 1672–1676, 2001.
[17] R. Vaughn and G. Evron, ―DNS amplification attacks,‖ 2006. [Online].
Available: http://www.isotf.org/news/DNS-Amplification-Attacks.pdf
[18] V. Paxson, ―An analysis of using reflectors for distributed denial-
ofservice attacks,‖ ACM SIGCOMM Computer Communication Review,
vol. 31, no. 3, pp. 38–47, 2001.
[19] F. Baker and P. Savola, ―Ingress Filtering for Multihomed Networks,‖
RFC 3704 (Best Current Practice), Internet Engineering Task Force,
Mar. 2004. [Online]. Available: http://www.ietf.org/rfc/rfc3704.txt
[20] C. Herley, ―So long, and no thanks for the externalities: The rational
rejection of security advice by users,‖ in Proceedings of the New
Security Paradigms Workshop (NSPW), 2009.
[21] J. Sunshine, S. Egelman, H. Almuhimedi, N. Atri, , and L. Cranor,
―Crying wolf: An empirical study of ssl warning effectiveness,‖ in
Proceedings of the 18th USENIX Security Symposium, 2009.
[22] R.Wagner, ―Address resolution protocol spoofing and man-in-the-
middle attacks,‖ The SANS Institute, 2001.
[23] A. Kapela and A. Pilosov, ―Stealing the internet - a routed, wide-area,
man-in-the-middle attack,‖ in Defcon 16, Las Vegas, August 2008.
[24] R. Kuhn, K. Sriram, and D. Montgomery, ―Border gateway protocol
security,‖ July 2007.
[25] R. Dhamija, J. Tygar, and M. Hearst, ―Why phishing works,‖ in
Proceedings of the SIGCHI conference on Human Factors in
computing systems. ACM, 2006, p. 590.
[26] D. J. Bernstein, ―DNSCurve: Usable security for DNS.‖ [Online].
Available: http://dsncurve.org
[27] T. Dierks and E. Rescorla, ―The Transport Layer Security (TLS)
Protocol Version 1.2,‖ RFC 5246 (Proposed Standard), Internet
Engineering Task Force, Aug. 2008. [Online]. Available: http:
//www.ietf.org/rfc/rfc5246.txt
[28] T. Ylonen and C. Lonvick, ―The Secure Shell (SSH) Protocol
Architecture,‖ RFC 4251 (Proposed Standard), Internet Engineering
Task Force, Jan. 2006. [Online]. Available:
http://www.ietf.org/rfc/rfc4251.txt
[29] E. Felten, ―Mozilla debates whether to trust chinese CA,‖ February
2010. [Online]. Available: http://www.freedom-to-tinker.
com/blog/felten/mozilla-debates-whether-trust-chinese-ca
[30] J. Corbet, ―China internet network information center accepted as a
mozilla root CA,‖ Linux Weekly News, February 2 2010. [Online].
Available: http://lwn.net/Articles/372264/
[31] P. Vixie, ―What DNS is not,‖ Communications of the ACM, vol. 50, no.
12, pp. 43–47, December 2009.
[32] M. S. Mimoso, ―Kaminsky interview: DNSSEC addresses cross
organizational trust and security,‖ June 2009. [Online]. Available:
http://searchsecurity.techtarget.com/news/interview/0, 289202,sid14
gci1360143,00.html
[33] T. Maly, ―I have some opinions about the RWW facebook login
hilarity,‖ February 2010. [Online]. Available:
http://quietbabylon.posterous.com/ i-have-some-opinions-about-the-
rww-facebook-l
[34] J. Sobey, R. Biddle, P. C. van Oorschot, and A. S. Patrick, ―Exploring
user reactions to new browser cues for extended validation certificates,‖
in Proceedings of the 13th European Symposium on Research in
Computer Security. Springer, 2008, pp. 411–427.
Invited Talk: Big Allied and Dangerous
– Terrorist Networks
Victor Asal and R. Karl Rethemeyer University at Albany, SUNY
1400 Washington Ave. Albany, NY 12222
[email protected], [email protected]
HY are some terrorist organizations so much more deadly than others? Why do some target the
United States? Why do some organization use or pursue CBRN? This presentation examines
organizational characteristics such as ideology, size, age, state sponsorship, alliance connections, and
control of territory to see how they can help us understand terrorist organizational behavior.
W
Abstract—Information warfare refers to actions that deny,
exploit, corrupt, or destroy the enemy’s information and its
functions. A model in this context classifies strategies for
attacking the enemy’s information infrastructure. Recent
research has defined four canonical strategies using Shannon’s
information theory. This paper introduces a complementary
approach that constructs a conceptual model of information flow.
The model is used to identify attack strategies.
Index Terms—Information warfare, information
infrastructure, attack strategies
I. INTRODUCTION
HE Information Revolution, manifested as an information
explosion aided by computers, mass media, and
telecommunication networks, has contributed to the concept of
warfare in relation to exploiting/denying information.
The importance of IW/IO [Information Warfare/
Information Operation] today is a simple consequence
of the reality that the digital infrastructure in the
developed world processes, concentrates and carries the
flow of knowledge which provides us with our decisive
economic and military advantage against other
players… The digital infrastructure is thus a lucrative
target, in every sense, since it provides a single point of
failure for the developed world‘s economic and military
power base. It is also a capable weapon for penetrating
our economic, military and political fabric, exploitable
by foreign and domestic players with hostile agendas.
[10]
Information warfare as a discipline deals with many issues,
including human factors, legal problems, rules of engagement,
damage assessment problems, and structuring forces to execute
information warfare (attributed to Winn Schwartau) [10]. An
important field of study in IW is the ways in which information
is exploited/protected.
In this paper, we analyze IW in terms of strategies of attack.
While such a contribution offers a new approach to the issue of
attack technicalities, the broader aim is to propose building a
foundation for IW upon the notion of flow of information.
Flow of information underlies all information systems, and this
is the reason we concentrate in section II on reviewing the
definition of IW and its notions of denying, exploiting,
corrupting, or destroying information. In section III, we
review a model for information in IW based on Shannon‘s
communication model. The purpose is outlined in section IV,
where the need for a conceptual foundation for IW analysis is
shown. Section V introduces such a conceptual foundation
through a model of information flow. Section VI applies the
concept of information flow to the problem of modeling
(information) combat theater. Specifically, this methodology is
used to analyze attack strategies.
II. DEFINITION OF INFORMATION WARFARE
According to Borden [5], the USAF model [16] and the
Network-Centric model [14] of Information Warfare define
IW as "any action to Deny, Exploit, Corrupt or Destroy the
enemy‘s information and its functions; protecting ourselves
against those actions and exploiting our own military
information functions" [16] (italics added).
We can investigate this focus on the operations of denying,
exploiting, corrupting, and destroying the enemy‘s information
as follows.
A. Deny the Enemy’s Information and its Functions
Smith‘s [17] description of this operation is ―IO can be used
to deny the enemy access to information. This may be done
through techniques like electronic warfare or the destruction of
command units‖ (italics added). So it seems that to deny
information means to prevent access to information about one
of the sides involved in a conflict. Fig. 1 illustrates information
denial/prevention. Denying information in this context seems
to denote a ―blockage‖ in information flow from our
information sphere to that of the enemy.
Flow-based Information Warfare Model
Sabah Al-Fedaghi Computer Engineering Department, Kuwait University
PO Box 5969 Safat, Kuwait 13060
T
A more complete view to be introduced in our conceptual
model is ―blockage‖ of the enemy from collecting/receiving
any type of information (Fig. 1(d)). Adopting such a concept,
we can concentrate on different types of blocked information:
our information, information about us, relevant information,
and information per se. The information warfare model ought
to incorporate such a concept of ―information suffocation‖ in
the definition of IW.
Thus, the use of ―deny‖ seems not to reflect any significant
operation on information. Collecting/receiving information is
an operation resulting in a basic state of information.
―Blockage‖ in this case refers to preventing information from
being collected/received information. This description, in
terms of states of information, will become clear when we
introduce our information flow model. The point here is that to
―deny information‖ is not a fundamental notion ―about‖ that
information.
B. Exploit the Enemy’s Information
Exploit refers to making the most of the enemy‘s
information. The verb exploit is used in the senses, ―use or
manipulate to one's advantage,‖ and ―make good use of.‖ The
US DoD views exploitation as an ―attack measure‖ defined ―as
gathering an adversary‘s flow of information to facilitate
friendly information generation‖ [5]. According to Borden [5],
since exploitation ―does not in itself produce an immediate
causal effect in the function of the target [communication]
channel, it cannot be classified as an offensive Information
Warfare strategy … Rather, it is an information gathering
‗technique‘.‖ The new IO (Information Operation) USA DoD
doctrine [12] does not consider ―Exploiting an enemy
information asset for intelligence purposes‖ an aspect of IO.
In addition, ―exploiting information‖ does not complement
―denying information.‖ ―Denying information‖ may be
regarded as a by-product of ―exploiting information.‖ For
example, according to Kopp [10], ―IW/IO are all operations . .
. conducted to exploit information to gain an advantage over
an opponent, and to deny the opponent information which
could be used to an advantage.‖ Thus, it seems that the terms
deny and exploit in the definition of IW do not reflect
systematic operations on information.
C. Corrupt or Destroy the Enemy’s Information
Corrupting information refers to changing information,
which creates new information (e.g., from Tanks approach
from south to Tanks approach from north). As we will see, this
new information is created from existing information. A more
complete description of IW thus includes ―the enemy‘s ability
to create information (from scratch or from old information).‖
We will view information flow in terms of five stages:
receiving, processing, creating, releasing, and transferring of
information. Creating information includes the notion of
corrupting information.
Destroying the enemy‘s information is one possible way to
disturb the enemy‘s information system. Other methods
include disturbing the internal flow of information such as ―not
processing received information,‖ ―not drawing a conclusion
(creating) when processing information,‖ etc. Even ―copying
(stealing) enemy‘s information‖ is an important operation that
might be more valuable than destroying that information.
We can see that Deny, Exploit, Corrupt, or Destroy the
enemy‘s information seems to be an incomplete description of
operations involved in IW. In section V, we describe an
information flow model that can be used to develop an
alternative specification of IW. In the next two sections, we
review and scrutinize IW modeling.
III. REVIEW: IW MODELS
Recently, the field of information warfare modeling has
been an active research area, e.g., [5][6][9][18]. Related works
have applied these models in several directions, including
information assurance and the semantic web. It seems that
none of these models addresses a high-level conceptualization
of the notion of information warfare. Every model typically
shows certain properties representing different viewpoints and
processes that occur during warfare-related processes. For
example, CycSecure Architecture introduced in [18] includes
such notions as ―Dynamic HTML User Interfaces,‖ CERT,
web-browser, etc. While these levels of detail are suitable for
the purposes of the authors, a conceptual description of
information warfare is different. It aims at identifying the
phases of a warfare life cycle via abstract and generic
(b) Prevent receiving information about
us
Enemy
Relevant
Information
(c) Prevent receiving information
Enemy
Information
about Us
Enemy
Our
Information
(a) Deny access to our information
Enemy
Information
(d) Prevent receiving information
Fig. 1. Difference between Deny and Prevent
receiving information
concepts.
A conceptual model is a high-level abstraction that
encapsulates main knowledge and behavior in relation to real-
world entities or phenomena. The model provides some
critical insights into the real world to be used in understanding
it and communicating about it.
Borden [5] and Kopp [9][10] proposed a model for
―information‖ in IW based on Shannon‘s model [15] for a
communication channel:
Consider for instance the radar detection model, where
a radar illuminates a target to establish its position,
kinematic parameters and possibly even identity. In this
model the ‗message‘ is encoded in the modulation and
timing of the power backscattered from the target, while
the ‗transmitter‘ is the reflecting body of the target
vehicle, which is supplied with power by the microwave
source in the radar producing the illumination. In this
example the attacker may degrade the S/N of the radar
by applying a microwave absorbent coating to his
vehicle, and in doing so hide the vehicle‘s radar return
in the background noise and noise floor of the radar‘s
receiver. [5] (Italics added).
According to Borden [5],
Information Warfare in the most fundamental sense
amounts to manipulation of a channel carrying
information in order to achieve a specific aim in
relation to an opponent engaged in a survival conflict.
[5, 10] (Italics added).
Notice that ―misinformation‖ in information theory is
related to ―noise‖ (unintentionally) produced by the channel.
In this context, misinformation is a type of information. Of
course this is not applicable to ―semantic information‖, where
the notion of a motive is applied to misinformation.
In view of Shannon‘s channel capacity theorem,
C = W Log2 (1+ P/N)
an attacker manipulates the flow of information by controlling
the capacity of the channel, using bandwidth, W, and signal
power vs. noise power, P/N [8].
Accordingly, four canonical strategies are available to
attackers (see Fig. 2) [8]:
- Degradation, or Denial, achieved by injecting noise and thus
degrading the channel. This is the so-called (covert) active
form of degradation, exemplified by injecting noise into the
channel to degrade capacity. In its passive form, the power in
the channel is driven to zero, burying the message carrying
information in background noise [8].
- Corruption, or Mimicry involves the attacker using a (covert)
signal that mimics the victim‘s signal so well it cannot be
distinguished from the real signal and is used instead.
- Denial, or Destruction involves the expedient of damaging
or destroying the enemy‘s receiver or the transmission link.
- Denial via Subversion or Subversion is an attack that
penetrates the internal functions of the victim to compel it to
do something self-destructive, using its own resources to
damage itself (e.g., virus and worm).
These strategies can be broken down into a collection
of more basic attacks, often mutually dependent but
each comprising only one of the four canonical
strategies. From a science perspective these strategies
can be said to be atomic in the sense that they are
simplest forms, from which all other more complex
forms are constructed. It is one of the realities of nature
that something of extreme complexity, such as strategic
deceptions, can be broken down into an interconnected
web of mutually dependent but essentially simple forms
of attack. What the four canonical strategies
demonstrate is that the vast footprint of various IW
techniques can all be broken down and modelled
mathematically, or in simulations. [11]
IV. SCRUTINIZING IW MODELS
Modeling of communication is an evolutionary process in
which new concepts enhance and complement earlier
communication models. Shannon‘s model is fundamental for
all subsequent communication models.
Consider the contentious point related to the interiority of
the channel. Conceptually, a disturbing element in Borden-
Kopp‘s model is the focus on channel capacity to classify
―basic attacks.‖ Implicitly, the model indicates that
information passes, or is transferred, through the presence of a
channel in the model. Such a view is analogous to
conceptualizing the notion of travel as transfer from one point
and arrival at another point. In a more comprehensive view,
the process of (air) travel includes the notions of being
received at the travel station (airport), processed (e.g., luggage
and passports), released to boarding (waiting for boarding),
and actual transfer onto the plane. On the other end, after
being transferred, passengers arrive, are processed, released,
and then transfer (leave for hotels). Similarly, information in
the communication stream is not just sent and received, but
also has repeated lifecycle states: it is received, processed
(changes its form), created (generates information from
information), released (e.g., waits for channel to be available),
and transferred.
We contend that the life cycle of information in any
communication system consists of iterations of stages
according to a state diagram. The five stages are receiving,
processing, creation, release, and transfer. Life cycle refers to
the ―birth‖ of a communicative act through initiation of a flow
of information directed to a certain destination, and the
―decay‖ of such an act through seizure of flow of information.
Seizure or stoppage of flow of information can occur at any
point in the stream of information flow regardless whether a
destination is reached. The flow stream comprises successive
stages of the stages described previously across different
ATTACKER’S INFORMATION SOURCE
ATTACKER’S TRANSMITTER
ATTACKER’S SIGNAL BURIED IN NOISE
RECEIVER DESTINATION NOISE SOURCE
(a) Degradation Strategy − Passive Form
ATTACKER’S NFORMATION SOURCE
ATTACKER’S TRANSMITTER
ATTACKER’S SIGNAL BURIED IN NOISE
RECEIVER DESTINATION NOISE SOURCE
Denial Strategy
ATTACKER’S NFORMATION SOURCE
ATTACKER’S TRANSMITTER
ATTACKER’S SIGNAL BURIED IN NOISE
RECEIVER DESTINATION NOISE SOURCE
(d) SUB/Denial Strategy
ATTACKER’S TRANSMITTER
ATTACKER’S INFORMATION SOURCE
ATTACKER’S NFORMATION SOURCE
ATTACKER’S TRANSMITTER
ATTACKER’S SIGNAL BURIED IN NOISE
Fig. 2. The four canonical strategies (from [11]).
DESTINATION NOISE SOURCE
(b) Degradation Strategy − Active Form
ATTACKER’S NOISE SOURCE
ATTACKER’S NFORMATION SOURCE
ATTACKER’S TRANSMITTER
ATTACKER’S SIGNAL BURIED IN NOISE
RECEIVER DESTINATION NOISE SOURCE
(c) Corruption Strategy
ATTACKER’S TRANSMITTER
ATTACKER’S INFORMATION SOURCE
participants‘ boundaries.
In addition, Shannon‘s model does not reflect conceptually
the ontological nature of communication to be taken as a
foundation for ―strategies that can be said to be atomic in the
sense that they are simplest forms, from which all other more
complex forms are constructed.‖
For example, it is well known that a communication act
involves information, a message (symbolic representation),
and signals (e.g., physical or electronic signals). The flow of
these three types of things is represented in Shannon‘s model
by a single arrow between the sender and receiver. Such a
conceptualization is analogous to representing gas, water, and
electricity lines in the design of a building with only one type
of arrow in the blueprint. Information is usually created by the
sender, while (signal) noise is created in the channel. The
noise is physical signals. Conceptually, this noise ought not to
be mixed with randomness (entropy) created at the source in
some applications. Entropy is a ―type of information,‖ whereas
noise generated in the channel comprises physical signals. Of
course, noise interweaves with messages while being
converted to signals for transmission purposes.
The Borden-Kopp model is limited by limiting the
operations of Corruption and Degradation to the channel
instead of along the line of flow of information. The channel
and its capacity are the centerpiece of operations, but the
model is inconsistent because it shifts direction in ―Denial via
Subversion or Subversion attacks‖ to non-channel operation
on the internal functions of the target.
From an evolutionary perspective on idealized
communication models, communication models evolved from
Shannon‘s model to a network model (see Fig. 3) that included
many details such as authentication, routing identification,
governing, data compression and decompression, and
detection of errors in transmission and arranging for their
correction. Hence, focusing modeling attacks on channel
capacity would seem to hinder development of a more
comprehensive approach to IW techniques.
Before such an approach is introduced, the next section
reviews the basic model used to explore an alternative
methodology for identifying critical security vulnerabilities in
information infrastructure.
V. THE FLOW MODEL
This section reviews the basic aspects of the Flow Model
(FM) to make the paper self-contained [1][2][3][4]. To
simplify this review of FM, we introduce flow in terms of
information flow.
Information passes through a sequence of states as it moves
through stages of its life cycle, as follows:
1. Information is received (i.e., it arrives at a new sphere,
similar to passengers arriving at an airport).
2. Information is processed (i.e., it is subjected to some type of
process, e.g., compressed, translated, mined).
3. Information is disclosed/released (i.e., it is designated as
released information, ready to move outside the current sphere,
such as passengers ready to depart from an airport).
4. Information is transferred to another sphere (e.g., from a
customer‘s sphere to a retailer‘s sphere).
5. Information is created (i.e., it is generated as a new piece of
information using different methods such as data mining).
6. Information is used (i.e., it is utilized in some action,
analogous to police rushing to a criminal‘s hideout after
receiving an informant‘s tip). Using information indicates
directing or diverting the information flow to another type of
flow such as actions. We call this point a gateway in the flow.
7. Information is stored. Thus, it remains in a stable state
without change until it is brought back to the stream of flow.
8. Information is destroyed.
The first five states of information form the main stages
of the stream of flow, as illustrated in Fig. 4. When
information is stored, it is in a substate because it exists in
different stages: information that is created (stored created
information), processed (stored processed information), and
received (stored received/row information).
The term ―information‖ is utilized in its common meaning.
It is a discrete unit (called flowthing [4]) that is received,
processed, created, released, and transferred, as described in
Fig. 4. The model can be redefined as a "signal flow model",
and also as other flowthings, such as money, materials, and
even actions, and concepts. A concept (whatever its nature) is
a ―thing‖ that can be received, processed, created, released,
and transferred. Accordingly, use of the term ―information‖ (a
signal, or content of a message) does not introduce difficulty at
a fundamental level.
Communication network
Messages
Recipients Senders
…
…
Fig. 3. Model of communication network (modified
from [13]).
Fig. 4. Information states in FM.
Created
Received
Released Processed Transfer
We use the term ―states of information‖ in the sense of
properties; for example, states/phases of water in physics:
liquid, ice, and gas.
The five-stage scheme can be applied to humans and to
organizations. It is reusable because a copy is assigned to each
agent or entity. Consider an information sphere that includes a
small organization with two departments; it comprises three
information schemes: one for the organization at large, and
one for each department. Fig. 5 shows the internal information
flow in such a sphere.
The five information states are the only possible ―existence‖
patterns in the stream of information. To follow information as
it moves along paths, we can start at any point in the stream.
Suppose that information enters the processing stage, where it
is subjected to some process. The following are ultimate
possibilities:
1. It is stored.
2. It is destroyed.
3. It is disclosed and transferred to another sphere.
4. It is processed in such a way that it generates new
information (e.g., statistical analysis generates the information
that Smith is a risk).
5. It triggers another type of flow. For example, upon receiving
patient health information, a physician takes some action such
as performing medical treatment. Actions can also be received,
processed, created, released, and communicated.
In Fig. 4, note that the arrows between Released on one
hand and the stages of Received, Processed, and Created are
bidirectional. This flow in opposite directions accounts for
cases when it is not possible to communicate information, as in
the case of a broken channel. In this case, at the Release stage,
the information can be:
- destroyed after a certain period
- stored indefinitely
- returned to release in the receiving, processing, or creation
stages
VI. MODELING COMBAT THEATER
FM assumes that the involved parties are represented by the
five components or stages of FM: receiving, processing,
creation, release, and transfer. Next, we examine these stages
in view of Shannon‘s communication model.
According to Shannon, the elements involved in
communication of information are source, transmitter, channel,
receiver, and destination.
A. Source/Transmitter
The source produces messages to be communicated to the
receiving terminal. FM extends this side of communication to
highlight the ―origin‖ of the message, whether received from
outside the source, or created within the source; thus, the
source can be described as creator or recipient, in addition to
being the sender of the message.
Fig. 5. Information flow within a company and its two departments.
Created
Received
Disclosed Processed Transfer
Created
Received
Disclosed Processed Transfer
Created
Received Disclosed
Processed
Transfer
Such a qualification may be significant in certain
circumstances (e.g., networking where communication
involves a chain of two-party exchanges).
The transmitter converts the message to a signal suitable for
transmission over the channel. In FM, the source has two
spheres, the messages sphere, and the signal sphere; thus, this
element of Shannon‘s model reflects the source as a processor
triggering the creation of signals that are released and
transmitted. This conceptualization distinguishes between two
explicit flowthings: information and signals, thus separating
the flow of information from the flow of signals.
Shannon‘s information theory makes a clear distinction
between signals and information. In many communication
systems, signals transmission is involved only in transferring
data, without the direct intent of conveying information along
with the data. In conventional terminology, the notion of data
is introduced as a form of information more suitable for
transmission. Considering data from the FM point of view,
data are processed (stage in FM) as digitally encoded
information. We thus have two ways to conceptualize the
relationship between information and data. If data are viewed
as a different flowthing from information, then another sphere
(besides information and signal) for data is distinguished in
FM; however, in a communication context, without loss of
generality, we view data as a form of processed information.
B. Receiver/Destination
Shannon‘s model abstracts transmission as a component of
the source and receiving as a component of the destination.
This is a reasonable way to look at the communication process,
because ―transmission‖ requires a deliberate act of message
releasing, and ―receiving‖ requires another deliberate act of
accepting the message. Even if ―transmission‖ is
conceptualized as being in the channel proper, the notions of
―transmission‖ and ―receiving‖ are still decisions to be made at
the source and the destination, respectively.
For example, it is possible that a message (e.g., an e-mail)
arrives at its destination; however, the communication process
is not completed if the receiver deletes it without reading it.
One objection to Shannon‘s model is that ―the receiver is
constantly being fed pieces of information with no choice in
the matter—willingly or unwillingly‖ [7]. FM is a more
suitable conceptual representation since it divides
communication into two types: one under the control of the
sender or receiver and one in the channel.
C. Channel
If the ―transmission channel‖ carries the signal from its
―transmitter‖ to a ―receiver‖ (e.g., device), this physical
activity is different from the abstract pre/post transfer stages of
―releasing‖ and ―receiving‖ the message at the source and
destination spheres. Furthermore, the nature of the channel is
different from the ―nature‖ of the source and the destination.
Clearly, the source and the destination deal first with
information, whereas the channel is a ―physical sphere‖ that
deals (in this case) with physical signals. Therefore, the basic
thing that is flowing (―transmitted‖ and ―received‖ in the
source and destination, respectively) is information, while the
thing that flows in the channel is only a physical signal. The
message has informational form when released by the source
and prior to channel transmission, and it returns to such a form
after channel transmission, when it is received at its
destination.
We propose to completely separate these spheres, as shown
in Fig. 6. The sender, channel, and receiver each have five
stages in the FM. The dotted arrows in the figure are
recognition that information/data flow is distinguished from
signal flow.
Adding different stages to the channel invites different
possibilities to be explored. For example, the channel is not an
implicit participant in the communication act; rather it is fully
represented as a communication sphere of signals flow. It
receives, processes, creates, releases, and communicates
signals.
Tran
sfer
Creation
Processing Releasing
Receiving
Tran
sfer
Creation
Processing Releasing
Receiving
Tran
sfer
Creation
Releasin
g
Processing
Receiving
Tran
sfer
Transfer
Fig. 6. FM version of communication.
Sig
nals
Sig
nals
Sender/
Source
Channel
Receiver/
destination
Receiving and communicating are the standard functions of
channels in current communication models. Creation in the
channel is manifested by noise (a type of ―signal‖) generated
in the channel. Noise is explicitly recognized in the channel.
The channel may ―delay‖ delivering the signal (e.g., traffic
congestion), thus putting it in the released state.
VII. MODEL FOR ATTACK STRATEGY
At this point, it is not difficult to develop a map of
information flow and to identify points vulnerable to attack. To
simplify the map, we follow the Borden-Kopp model and limit
the representation to a theater combat model involving three
elements: attacker, channel of communication, and victim, as
shown in Fig. 7. In this simplified conceptualization, if the
information area involves a network with multiple nodes
separating the attacker from the victim, the entire network is
represented as a channel.
A theater here refers to the informational area (sphere) of
action of conflict. This area generally consists of information
spheres and subspheres (e.g., victim‘s sphere, which includes
victim‘s subdivisions; see Fig. 5) represented as FM schema
including flows, and triggering relationships. The theater
appears as a map of information spheres and their information-
relevant flows.
Fig. 7 shows various information spheres for the attacker,
channel, and victim, and sample attacks. Attack 1 (Information
noise) represents what the Borden-Kopp model calls
―Degradation Strategy − Passive Form‖ in Fig. 2(a). The
attacker creates ―information noise,‖ releases it, and transfers
it to the channel.
Semantically, information noise is different from signal
noise generated inside the channel. Spamming is an example
of such a noise, where the receiver is overwhelmed with too
many messages. This type of attack is different from jamming
the content of a message with signals/symbols that distort the
original content.
In the channel, noise can be triggered (attack 2 – signal
noise) by different methods, including damaging the channel
physically, directly injecting a signal, etc. Jamming, for
example, involves (deliberate) transmission of signals that
disrupt communications by decreasing the noise ratio.
However, in this paper we are not concerned with multi-
dimensional consequences such as that jamming may alert the
victim that he is under attack.
The information spheres of the victim can be targeted by
many types of attacks:
- Victim‘s receiving system can be damaged
- After receiving a message, victim‘s receiving system can
malfunction or be starved for storage space and overwhelmed
by a shortage of power, parts, speed, etc.
- The value of received data can be devalued by denying the
ability to process it
- Processed data can be devalued by preventing analysis that
draws conclusions (new knowledge) because of lack of
processing power, speed, necessary memory storage, etc.
- Created data (conclusions) can be prevented from being
released/transferred to decision makers.
In comparison with the forms of the Borden-Kopp model,
these strategies can be said to be atomic in the sense that they
are the simplest forms and those from which all more complex
forms are constructed. Attacks can follow the enemy‘s
information, with one attack after another along the
information flow, like bullets chasing a target from one place
to another.
VIII. CONCLUSION
This paper introduces a new approach to IW that constructs
a conceptual model of information flow. The model is used to
identify attacking strategies. IW can be defined in terms of the
resultant model, such as IW undertaken to undermine the
information sphere of an enemy. Undermining an information
sphere refers to causing malfunctioning states and flows of
information.
Tran
sfer
Creation
Processing Releasing
Receiving
Tran
sfer
Creation
Processing Releasing
Receiving
Tran
sfer
Creation
Releasing Processing
Receiving
Tran
sfer
Transfer
Fig. 7. FM version of sample attacks.
Sig
nals
Sig
nals
Sender/
Source
Channel
Receiver/
destination
1. Information noise 2. Signal noise
The flow-based model does not deal with the content of
information. The type of information can be studied at the next
level. Our model is a general conceptualization of flow of
information (receive information, transfer information, etc.). It
is at the same level as conceptual models for understanding
knowledge that includes such phases as knowledge scanning,
knowledge evaluation, knowledge transfer, and knowledge
application without specifying what knowledge. It is also
similar to the famous observe-orient-decide-act (OODA)
model of attack, where the types of actions are not specified.
REFERENCES
[1] S. Al-Fedaghi and A. Alrashed, ―Threat risk modeling,‖ presented at the
International Conference on Communication Software and Networks
(ICCSN 2010), Singapore, February 26–28, 2010.
[2] S. Al-Fedaghi, ―How the Pride Attacks‖, presented at the 9th European
Conference on Information Warfare and Security, Thessaloniki, Greece,
1-2 July, 2010.
[3] S. Al-Fedaghi, ―Modeling communication: One more piece falling into
place,‖ presented at The 26th ACM International Conference on Design
of Communication (SIGDOC 2008), Lisboa, Portugal, Sept. 22–24,
2008.
[4] S. Al-Fedaghi, ―Systems of things that flow,‖ presented at The 52nd
Annual Meeting of the International Society for the Systems Sciences
(ISSS 2008), University of Wisconsin, Madison, USA, July 13-18,
2008.
http://journals.isss.org/index.php/proceedings52nd/article/viewFile/100
1/354
[5] A. Borden, ―What is information warfare?‖ Aerospace Power
Chronicles, United States Air Force, Air University, Maxwell AFB
USAF, IWS: Information Warfare Site, November 2, 1999.
http://www.iwar.org.uk/iwar/resources/airchronicles/borden.htm
[6] L. S. Johnson, ―Toward a functional model of information warfare,‖
Central Intelligence Agency, May 8, 2007.
https://www.cia.gov/library/center-for-the-study-of-intelligence/kent-
csi/docs/v40i5a07p.htm
[7] M. McLuhan, Understanding Media: The Extensions of Man. New
York: The New American Library, 1964. Excerpt: ―Interpretations with
limitations: A critical essay on the mathematical theory of
communication‖ at http://www.uweb.ucsb.edu/~andreabautista/
[8] C. Kopp, ―The four canonical strategies of information conflict,‖
Lecture 04, 2006.
http://www.csse.monash.edu.au/courseware/cse468/2006/Lectures/CSE-
468-04.pdf
[9] C. Kopp, ―Shannon, hypergames and information warfare,‖ Journal of
Information Warfare, vol. 2, no. 2, 2003, pp. 108-118.
http://www.csse.monash.edu.au/courseware/cse468/2006/Lectures/_JIW
-2002-1-CK.pdf
[10] C. Kopp, ―Information warfare. Part 1. A fundamental paradigm of
infowar,‖ February 2000.
http://www.ausairpower.net/OSR-0200.html
[11] C. Kopp, ―Fundamentals of information warfare,‖ NCW 101 Part 14,
DefenceToday. http://www.ausairpower.net/NCW-101-14.pdf
[12] R. L. Mackey, ―Information operations: Reassessing doctrine and
organization,‖ 2003.
http://www.dtic.mil/cgi-
bin/GetTRDoc?AD=ADA413656&Location=U2&doc=GetTRDoc.pdf
[13] A. Pfitzmann and M. Hansen, ―Anonymity, unlinkability,
unobservability, pseudonymity, and identity management – A
consolidated proposal for terminology,‖ Aug. 25, 2005.
http://dud.inf.tu-dresden.de/Literatur_V1.shtml
[14] D. Silverberg (Ed.), Network centric warrior, Q&A. Interview with Vice
Admiral Arthur Cebrowski, Director, Space, Information Warfare,
Command and Control, Chief of Naval Operations. Military
Information Technology, vol. 2, no. 2, April-May 1998.
[15] C. E. Shannon, ―A mathematical theory of communication.‖ The Bell
System Technical Journal, vol. 27, July, October 1948, pp. 379–423,
623–656.
[16] S. E. Widnall and R. R. Fogelman, "Cornerstones of information
warfare," SAF/CSAF publication, 1997.
http://knxas1.hsdl.org/homesec/docs/dod/nps08-021105-01.pdf
[17] R. Smith, ―Simulating information warfare using the HLA Management
Object Model,‖ SIW Discussions, Spring 2000.
http://www.modelbenders.com/papers/00F-SIW-021.pdf
[18] B. Shepard, C. Matuszek, C. B. Fraser, W. Wechtenhiser, D. Crabbe, Z.
Güngördü, J. Jantos, T. Hughes, L. Lefkowitz, M. J. Witbrock, and D.
Lenat, ―A knowledge-based approach to network security: Applying Cyc
in the domain of network risk assessment,‖ in Proceedings of the
Seventeenth Innovative Applications of Artificial Intelligence
Conference, Pittsburgh, Pennsylvania, July 2005, pp. 1563-1568
A Multi-Staged Approach to Risk
Management in IT Software Implementation
Sanjukta Das, Raj Sharman, Manish Gupta, Varun N. Kutty, Yingying Kang
State University of New York, Buffalo, NY
Amherst, New York, USA, 14260
{sdsmith4, rsharman, mgupta3, vn27, ykang4 }@buffalo.edu
Abstract: Software implementations in organizations are
highly complex undertakings with non-trivial risk implications
pertaining to data, applications and other resources. This
research in progress paper presents two-tiered approach to
managing risk over a multi-staged implementation. As an
example, we use the application area of Single Sign-on (SSO)
whereby a user can gain access to multiple, interdependent
software systems using a single username/password
combination. We develop an optimization strategy that enables
addition of appropriate software systems to SSO
implementation in multiple phases. We formulate an integer
programming (IP) problem to determine the optimal cluster of
applications to be moved to SSO in each phase based on the
inherent risks involved therein. We then use Real Options
Theory to help managers determine the timing of the different
stages of their investment decision.
I. INTRODUCTION
HIS research in progress paper addresses issues with
IT software implementations in organizations that
tend to increase in complexity with greater amounts
of pervasiveness, data complexities and other such factors.
Such implementations can be fraught with a variety of risk
implications pertaining to a range of resources. We offer a
framework for managing risks in a multi-staged
implementation environment pertaining to an exemplar
application area of Single Sign On (SSO). Due to
proliferation, SSO is an authentication mechanism by which
a user can gain access to multiple, software systems and
organizational resources using a single username/password
combination. Without SSO, users have to manage a set of
authentication credentials for every system they need to
access. A single sign on system should provide for secure
storage of user passwords, support for more than one user
password, as well as support for multiple target logon
methods [3]. The research in the area of SSO is at best
scanty. Further, the investigation of implementation issues
from a research perspective and the application of real
options in SSO investments are virtually non-existent. This
paper develops an optimization strategy that provides
managerial insights on implementation options available to
managers using an IP formulation and real options theory.
In this paper we have assumed the software systems to be
independent of each other. The risk mitigation strategies in
SSO are of particular importance to every organization
because of the security concerns when adopting SSO.
The contributions of this paper are twofold; risk
mitigation strategy by adopting SSO and timely
implementation of the investment strategy through effective
managerial decision making by the use of real options. The
IP formulation solves the problem by provision of a solution
that informs on a risk mitigating strategy on which
applications to implement as part of the single sign-on in
each phase based on a number of constraints. Gordon et. al.
have also real-options based approach of “wait and see” in
security investments does make sense for most of the
information security capital expense decisions [5]. Real-
options analytic framework have been used to show that
information security investments are time based and are
sometimes deferred [7]. The decision to invest is made by
considering real options. Herath & Herath [6] developed an
integrated real options model for information security
investments using Bayesian statistics that incorporates
learning effects. The advantages of moving to SSO are a
reduction in time spent in entering passwords, improved
security by the reduced requirement of handling multiple
passwords, reduced helpdesk costs by lowering the number
of helpdesk calls related to password reset and centralized
reporting for compliance adherence. SSO also makes the
satisfying of audit requirements (such as compliance with
Sarbanes-Oxley (SOX)) easier, robust and less expensive as
the process can become more centrally managed. The
disadvantages of moving to SSO are single point of failure
and single source of attacks or hacking.
II. MODEL PARAMETERS
In this paper, we assume that application software
systems are first classified into several risk categories such
as high, medium or low risk or more specific classes such as
customer facing, employee facing, federal government
facing, etc symbolizing the risk they pose. Applications are
considered for inclusion in a phase for implementation by
taking into consideration costs and threshold data risks
involved per period based on security policies in place.
Threshold data risk (TDR) is a function of other
organization factors and requires analysis by IT experts to
the level of data risk they wish to assume per phase of
implementation (details on the computation of such risks is
beyond scope of this paper – we assume this is computed or
provided for the purposes of this paper).
Also, three matrices have been defined:
a. Application Confidentiality Matrix AC,
if application j contains
confidential data, else .
b. Application Data Size Matrix AD,
= data in Megabytes, handled
by application j.
T
c. We define a matrix
, to be used in the formulation. This
matrix specifies which data belonging to application j is
confidential.
Consider matrix for Total Variable Cost
, where is the number of users using
software system j at time t and is the variable cost
per user per application j at time t.z`
III. COST MINIMIZATION PROBLEM
The Integer Programming problem here is to identify
optimal cluster of software systems per phase to be adopted
for Single Sign On at the lowest cost. The problem takes
into account the data exposure risk and business criticality
risk involved and gives an optimal solution. At the outset
we define the notation used in the formulation in Table I.
Minimize:
Subject to:
(1a)
(1b)
(2a)
(2b)
(3)
(4)
(5)
(6)
The formulation is presented in equations 1a-6. The
objective function consists of fixed costs and variable costs
(licensing fees, training costs). Constraint (1a) takes care of
data threshold risk (TDR) for every implementation phase.
As the implementation team gains confidence in
implementing SSO, the learning effect is captured by the
=incremental values of the threshold risk that are assigned
to each phase. For example, TDR may be modeled as a
linear function of time or a non-linear function such as an
exponential function as is relevant to that organization.
Constraint (1b) ensures that some of individual risk
categories of the software systems are less than the data
threshold risk per risk type. Either (1a) or (1b) could be
TABLE I NOTATIONS
Symbol Description
FC0 Initial fixed cost
FCt Fixed cost (of implementation) at time interval t, assumed to remain constant per time period
TVCt,j Total Variable Cost using software system j at time t
to SSO at time interval t r(q)j This is risk stemming from all factors other than
data. For instance, a bank could have risk levels
related to customer/employee facing applications, federal filings, etc. Another firm might simply rank
risks as high, medium-high, medium, medium-low,
low, etc. In general, r(q)1,j has a value of 1 if software system j is of risk type r(q) and 0
otherwise.
TVCt,j Total Variable Cost using software system j at time t to SSO at time interval t
r(q)j This is risk stemming from all factors other than
data. For instance, a bank could have risk levels
related to customer/employee facing applications,
federal filings, etc. Another firm might simply rank
risks as high, medium-high, medium, medium-low, low, etc. In general, r(q)1,j has a value of 1 if
software system j is of risk type r(q) and 0
otherwise. AppDataRiskj Risk stemming from confidential data that the
application supports TDRt Threshold Data Risk at time interval t
n Number of software systems
T Total time horizon R Total types of software system risks
LowerTper Implement each high risk software system every
Tperiodic number of years UpperTper Implement ( Tperiodic )
-1 high risk software systems
every year
Bt Maximum Budget at time interval t xjt Binary variable indicating whether software system
j was moved to SSO at time interval t
used based on management preferences. This is done for
security purposes as we wish to limit data exposure to a
minimum at every phase. Either constraint (2a) or (2b),
which is meant to cluster specific software systems per
phase, is used based on the following decision:
Let us calculate
There are two scenarios to be considered when clustering
the r(q) type of risk software systems, only one of them
must be implemented based on value of Tperiodic(q):
Constraint 3(a): Tperiodic (q) > 1: This means that we do
not have to implement r(q) risk systems every year. We can
implement one system every lower integer value of
Tperiodic(q) or LowerTper(q) number of years. For example,
suppose total number of periods is five years, number of
high risk systems is four. Therefore, LowerTper(q) = lower
integer bound (1.25) =1; which implies one high risk
software systems are implemented every year.
Constraint 3(b): Tperiodic (q) 1: We take the greater
integer value of 1/Tperiodic(q) or UpperTper(q).For example,
suppose total number of periods is 5 years, number of
customer facing applications (which is ranked fairly high on
the risk scale by the firm) is seventy. Therefore,
UpperTper(q)= upper integer bound (1/0.0714285) = 14;
which implies fourteen of these software systems are
implemented every year.
Constraint (4) assigns the lowest risk software systems in
every phase on the basis of costs per phase and assumable
risk per phase after taking into account all higher level risks
in Constraints (3) and (1). Once a software system has been
implemented under SSO it is no longer available for
reimplementation in another phase and this is ensured via
Constraints (4). Constraint (5) is used to restrain costs from
exceeding a budgetary threshold limit in every phase.
IV. REAL OPTIONS
Real Options Analysis is a method that allows
consideration for deferment among other possible actions in
the context of valuation of investments which traditional
budgeting techniques such as Net Present Value (NPV)
analysis cannot accommodate. Real Options computes the
value of embedded options (managerial flexibility)
separately and adds it to the traditional NPV, also called
Passive NPV (NPVP, which is obtained by subtracting cost
from utility gained, both present valued by a single period at
a risk free rate, which is the assumption made). What we
then get is the Active NPV (NPVA) [1][2].
We have defined the decision making process as such:
i. Invest if NPVP >0 and Option Value>0
ii. Abandon two tiered investment in first
phase if NPVA<0, NPVP<0 and Option Value>0
iii. Defer investment by one period if NPVA >
NPVP
iv. After first deferral, in the second phase,
we implement investment if NPVP>0 and Option
Value>0 or abandon if NPVP<0 and Option Value>0.
If the very first investment is abandoned we abandon
subsequent movements to SSO, because of sequential
consideration of implementation.
It is assumed that as more and more software systems are
added to SSO, employees become more familiar with the
system as time progresses and therefore, variable costs and
training costs reduce over time.
Fig. 1. Help desk cost per number of phases (n).
Help desk costs monotonically decrease as more software
systems are included under SSO. This is because of a
reduction in the number of password related calls. Helpdesk
costs is assumed to be an exponentially decreasing function
(Cost (n) = constant*e-k*(n-1)
), as shown in Figure 1. When
implemented for organizations, help desk costs will be a
function of number of users, help desk cost per user and
number of calls per user per year1. Upon deferment,
helpdesk costs in the next year are considered to be the
same as the previous year. An increase in user utility (or
productivity) is observed from the adoption of SSO, as users
spend a reduced amount of time making helpdesk related
calls and therefore is related to a reduction in help desk
calls. Utility function is assumed to be an increasing
function of nth
root of number of phases implemented
(Utility (n) = constant*(n1/k
)), as in Figure 2. The idea is that
utility increases as more and more software systems are
added to SSO.
Fig. 2. Utility over number of phases (n)
Satisfaction surveys such as SERVQUAL are adopted to
reach a consensus on the utility function needed to be
modeled. The observed increase in utility is a function of
percentage reduction in helpdesk costs, average wage per
user per annum, number of help desk calls per year
(assuming 40 hours work time per week for 52 weeks) and
time per help desk call.
Fig. 3. Technology cost over time
In Figure 3, it is seen that technology costs have also
been considered to be a monotonically decreasing
exponential function. We consider investment decisions two
phases at a time as in Figure 4.
Fig. 4. Investment decision tree
Figure 5 explains via pseudo code the method used to
apply real option managerial decision making.
Fig. 5. Real Options Algorithm
V. PROOF OF CONCEPT
Related work has been drawn from [4]. The paper
considered the case of a major North-East US Financial
Institution which considered moving to single sign on. The
parameters chosen to cluster software systems were user
base, criticality based on business classification,
authentication mechanisms, regulations applied to content
and process around application and application architecture.
Similar application characteristics and business aspects of
decision making are borrowed from that published work.
Ideas for research of this conference paper were kindled by
some tangential feedback we received from practitioners
working in the area during development of [4].
VI. CONCLUSION
In this paper, we develop a multi-staged SSO
implementation strategy that optimizes cost based on
risk thresholds using an Integer Programming
formulation and Real Options analysis. The IP
formulation tries to find the best clusters of software
systems along the time dimension by minimizing total
cost involved across all phases. The real options
analysis imparts managerial flexibility by determining
the ideal timing issues. This mechanism has been
implemented using CPLEX and MATLAB. A full-
fledged data collection and trace driven simulation will
be conducted next. This implementation does not take
into account typical organizational roles embodied in
Role Based Access Control (RBAC) and is an area for
future extension.
REFERENCES
[1] M. Benaroch, “Option-Based Management of Technology Investment Risk”, IEEE Transactions on Engineering Management,
Vol. 48, No. 4, pp. 428-444, November 2001.
[2] M. Benaroch, "Managing Information Technology Investment Risk: A Real Options Perspective," Journal of MIS, Vol. 19, No. 2, pp. 43-
84, Fall 2002.
[3] R. Cohen, R. A. Forsberg, P. A. Kallfelz, J. R. Meckstroth, C. J. Pascoe, A. L. Snow-Weaver, “Coordinating user target logons in a
single sign-on (SSO) environment”, Patent number: 6178511, Filing
date: Apr 30, 1998, Issue date: Jan 23, 2001, Assignee: International Business Machines Corporation.
[4] M. Gupta, R. Sharman, D. Bryan, “Enabling Business and Security
Through Technology Implementation: A Financial Services Case Study”. Journal of Applied Security Research, 1936-1629, Volume 4,
Issue 3, 2009, Pages 322 – 340, Taylor and Francis Group.
[5] L. Gordon, M. Loeb, and W. Lucyshyn, "Information security
expenditures and real options: A wait and see approach," Computer
Security Journal Volume XIX, Number 2, 2003, pp.17.
[6] H. Herath and T. Herath, “Investments in Information Security: A
Real Options Perspective with Bayesian Postaudit," Journal of
Management Information Systems, Winter 2008-9, pp. 337-375.
[7] K. Tamtumi and G. Makoto, "Optimal Timing of Information
Security Investment: A Real Options Approach" 2009 Workshop on Economics and Information Security (WEIS).
1. While number of phases is less than or equal to two
1.1. If NPVP>0 and Option Price>0 1.1.1.Implement Phase I
1.2. Endif
1.3. If NPVA>NPVP and NPVP<0 and Option Price>0 1.3.1. Defer Phase I
1.3.1.1. If NPVP < 0 and Option Price>0
1.3.1.1.1. Abandon Phase I 1.3.1.1.2. Go to 2.
1.3.1.2. Endif
1.3.1.3. If NPVP > 0 and Option Price>0 1.3.1.3.1. Implement Phase I
1.3.1.4. Endif
1.4. Endif 1.5. If NPVA<0 and NPVP<0 and Option Price>0
1.5.1. Abandon Phase I
1.5.2. Go to 2.
1.6. Endif
1.7. Increment phase number counter
2. Stop while loop
A Survivability Decision Model for Critical
Information Systems Based on Bayesian Network
Abstract—Critical information systems (CISs) cover vast
number of applications and are now an essential part of our day-
to-day life. Any damage to such a system or loss of information
due to malicious attacks or system failures can cause serious
consequences to society and individuals. Therefore, it is
important to maintain the survivability of the systems and make
timely decisions on system repair, if necessary, in order for the
systems to support critical services. In this paper, we propose a
Bayesian network based decision model to help system
administrators better understand the hidden states of a CIS in
order to determine its survivability status based on prior
knowledge and current available evidence. We perceive that the
survivability of a system is dependent on several factors in such a
way that probabilistic relationships exist between these factors
and the system’s survivability status. We represent such
probabilistic relationships using a Bayesian network. Our model
can be used to determine whether it is adequate for the system to
continue supporting critical services or it needs to be repaired to
avoid further losses. Therefore, the model helps system
administrators to reduce the magnitude of possible service
interruptions due to malicious attacks or system failures. 1
Index Terms—-Bayesian networks, survivability, decision-
making
I. INTRODUCTION
UR society is increasingly dependent upon large-scale,
highly distributed information systems. Numbers of
critical information systems (CISs) have been applied to
both civilian and military critical infrastructure applications
[1, 2] such as banking and finance, telecommunication,
healthcare, and national defense — which are essential for the
well-being of our society. In many cases, the provision of
service by the infrastructure applications depends on the
correct operations of the underlying critical information
systems; and damage or loss of service in any of these
application domains would cause serious consequences.
Nowadays, much attention has been paid to the dependence
on these critical information systems because of their ultimate
importance. This dependency results in a great deal of
concerns about those systems in case of hardware or software
failures, operator errors, power outage, environmental
disasters, and attack by adversaries [3].
Despite the best efforts of system developers and security
This work was supported in part by U.S. AFOSR under grant FA 9550-09-
1-0215.
practitioners, it is still infeasible to assure that a CIS will be
invulnerable to well-organized attacks and unpredictable
system failures. The unique characteristics of a CIS make it
infeasible to build a completely secure system. First, a CIS is
typically very large in size both geographically and in term of
the number of elements. Second, the services supplied to end
users are often attained by composing different
functionalities. Thus, the complete services can only be
obtained when several subsystems cooperate and operate in
some predefined order [2]. The cost of disruptions in CISs
grows more rapidly than ever before, yet increasing reliance
on CISs increases vulnerability to disruption. Given the
dependence upon these CISs, an important property that these
systems must achieve is survivability. In general, survivability
refers to the ability of a system to continue to provide service
in the face of failures, attacks, or accidents [4].
Given the importance of a CIS in various mission-critical
applications, it is crucial to constantly monitor the system,
detect any anomaly and conduct timely diagnosis and repair,
if necessary, to make sure that the systems can continuously
function and provide the essential services to users. However,
given the increasing complication of malicious attacks and the
increasing complexity of the systems, it is often difficult to
determine if a CIS has been compromised or in a faulty state
due to reasons such that the symptoms may not be observed at
an early stage. It is also challenging to determine whether a
CIS should be allowed to continuously function given its
uncertain state. On one hand, taking the system offline for
diagnosis and repair will unavoidably affect system
productivity. Therefore, such kinds of system check should
be done at a minimum frequency. In some situations, any
disruption of the system, even for a short period of time, may
not be accepted. On the other hand, it could result in
disastrous consequences if a CIS has already been
compromised and it still carries out mission- critical
functions. In this paper we propose a probability based
decision model to help system administrators to make the best
judgment on the current state of a CIS given various
influential factors and some known evidence about the system
properties. Our model helps users to determine whether it is
necessary to conduct a thorough system diagnosis and
whether it is acceptable for the system to continue supporting
mission- critical services. The model applies cause-effect
relationships among different factors and available evidences
Suhas Lande, Yanjun Zuo, and Malvika Pimple University of North Dakota
Grand Forks, ND 58203
{suhas.lande, yanjun.zuo, malvika.pimple}@und.edu
O
to reach the best actions for the system administrators to take
in order to maintain system survivability.
Bayesian networks are powerful tools for modeling cause-
effect relationships in various domains. They have proven to
be a useful tool in representing complex probability
relationships, such as determining the decision given evidence
from a set of dependent factors. In this paper, we apply the
Bayesian network to our decision model. Most of the security
related decisions are not momentary. They take time, are
dependent on other factors and can be divided into sub-stages
[5]. In our work the final decision of whether to allow a
system to continue supporting the essential services is
derived from a number of survivability influential factors
and the relationships among them.
The rest of the paper is organized as follows: Section 2
describes the related work; the survivability decision model is
presented in Sections 3; we conclude our paper in Section 4.
II. RELATED WORK
In this section, we discuss two streams of related work:
system survivability and applications of Bayesian network in
security and risk management.
A. System Survivability
Knight et al. [6] propose system architecture to enhance
the survivability of critical systems. In the paper they discuss
reconfigurable architecture in case of system errors and
propose a mechanism to achieve survivability. Such
architecture allows the detection and response to various types
of faults. By introducing fault tolerance, progress can be made
towards meeting the survivability goals of a critical
infrastructure application. Ellison et al. [7] highlight the
importance of survivability in critical systems. They present
survivability strategies to make a system robust and resilient to
attacks. The authors propose the use of traditional computer
security solutions, effective risk mitigation strategies based on
"what-if" scenario analyses and contingency planning to
improve the survivability of a system. Incorporating these
approaches into new and existing systems can improve their
survivability. Zhu et al. [8] present techniques of intrusion-
tolerance and multi-version subsystems for survivability of
critical information systems. The survivability goal is achieved
using redundancy and diversity in the presence of intrusions or
failures. Sullivan et al. [9] propose hierarchical, adaptive
distributed control systems to monitor and respond to
malicious attacks by changing the operating modes and
configurations of a system to minimize loss of system utility.
The authors evaluate their approach using distributed dynamic
models of infrastructure information systems.
In this paper we develop a model to determine the hidden
states of a system and decide whether the system is still
survivable to provide critical services in face of malicious
attacks and system failures based on intelligent evidences and
important factors that affect the survivability of the system.
The major goal of our work is to facilitate survivability
decisions using a probabilistic reasoning model.
B. Bayesian Network and its Applications
Bayesian networks are powerful tools for modeling cause-
effect relationships. They are compact networks of
probabilities that capture the relationships among events.
Formally, a Bayesian network is a joint probabilistic model for
n random variables (x1, . . . , xn) represented as a directed
acyclic graph G = (V, E) where V is a set of nodes
corresponding to the variables V = (x1, . . . , xn) and E V *
V contains directed edges connecting nodes in an acyclic
manner. Instead of weights, the graph edges are described by
conditional probabilities of nodes given their parents that are
used to construct a joint distribution P(V) or P(x1, . . . , xn).
Bayesian network has been applied to various applications
in security and risk management, where evidence and cause-
effect relationships can be used to predict probabilistic
unknown values of variables of interest. Modelo-Howard et al.
[10] propose an algorithm which systematically performs
Bayesian inference to determine the effect of intrusion
detector configuration on the accuracy and precision of
security goals in a system. The framework represents the
causality relationships between different attack steps and also
between attack steps and the detectors using conditional
probabilities. Cemerlic, et al. [11] propose an adaptive
network intrusion detection model using a Bayesian network,
trained with a mixed dataset containing real-world and
DARPA dataset traffic. Since individual events in an attack
can be represented as nodes and the causal relations among
events can be modeled as edges in Bayesian networks, their
paper uses a Bayesian network as the inference model.
Frigault, et al. [12] propose a Dynamic Bayesian networks-
based model to incorporate temporal factors, such as the
availability of exploit codes or patches. Starting from the
model, the authors study two concrete cases to demonstrate the
potential applications. The work provides a theoretical
foundation and a practical framework for continuously
measuring network security in a dynamic environment.
Bringas, et al. [13] present a unified misuse and anomaly
prevention system (ESIDE-Depian) based on Bayesian
networks to analyze network packets. The model is evaluated
against well-known and new attacks showing how it
outperforms a well-established industrial network intrusion
detection system. Abouzakhar, et al. [14] develop a
probabilistic approach using Bayesian Network to detect
distributed network attacks as early as possible. Their paper
shows how probabilistically Bayesian network detects
communication network attacks. Mukhopadhyay et al. [15]
propose a framework, based on copula aided Bayesian Belief
Network (BBN) model, to quantify the risk associated with
online business transactions, arising out of a security breach,
and thereby help in designing e-insurance products. A
representative causal diagram is developed elucidating the
probable reasons for a security breach and populated it with
expert opinions. Herath, et al. [16] develop a model for
information security investments using Bayesian inferences
for valuation and post-auditing. Their work uses Bayesian
statistics to model learning options and the post-auditing of
sequential real options, an idea that can be applied to a variety
of other types of sequential investment problems.
Our work has different objectives from the above frameworks and complements the existing models in the applications of Bayesian network in different areas. We use Bayesian network to represent the cause-effect relationships between various factors that affect the survivability of a system. Based on those relationships and user input intelligent evidence, systems administrators can determine the survivability status of a CIS. Those decisions are very essential to ensure the system can provide mission-critical services to users.
III. THE SURVIVABILITY DECISION MODEL
In this section, we discuss our Bayesian network based
survivability decision model. The Bayesian networks that can
be used strictly to model the reality are called belief nets,
while those that also mix elements of values and decision
making are called decision nets. The latter also provide
optimal decisions based on a set of evidences. In this paper,
we develop a Bayesian decision network model to be applied
to survivability related decision making. A Bayesian network
allows a user to enter the expert knowledge and available
evidences about a system’s status. Then the network can be
recalibrated every time when additional information is
entered. The effect of new information propagates
automatically to the children and parents nodes, enabling the
network to update the joint probability distributions. Bayesian
networks are easily extended to computing utility, given the
degree of knowledge we have on a given situation. They can
be used in any walk of life where modeling an uncertain
reality is involved, and, in the case of decision networks, it is
helpful to make intelligent, justifiable, quantifiable decisions
that will maximize the chances of a desirable outcome. Our
Bayesian network based survivability decision model helps
system administrator to choose the best decision by
computing the utility associated with each decision choice. In
our model, the utility is computed as the cost of interruptions
of service provided by a CIS. Since a CIS must survive
malicious attacks and system failures in order to provide
essential services to users whenever those services are
required, the time and efforts required to diagnosis or repair a
CIS should be at a minimum. More specifically, our
survivability decision model will address the following
issues:
Identifying the major factors that affect a CIS system's
survivability and specifying their quantitative cause-
effect relationships;
Defining the survivability decision objectives for a CIS,
a series of decisions for the system administrators to
make, the utilities of various choices for each decision,
and the conditional probability relationships (CPRs)
between the influencing nodes (i.e., the parent nodes)
and the influenced node (i.e., the child node);
Based on the specific CPRs and user input evidence,
answering various ―what-if‖ questions related to system
survivability which can be used by the system
administrators given different attack and system failures
scenarios.
Our model will assist administrator to better understand
the actual states of a CIS related to its survivability status.
This set of knowledge can help administrator to decide
whether to conduct an offline diagnosis of the system to fully
check the system’s survivability status. The diagnosis results
(with a certain confidence level) can further help the system
administrators to determine whether it is acceptable to allow
the system to continue supporting mission critical services or
they should stop the system to schedule a repair because the
system has been compromised. These decisions are critical in
maintaining a healthy/survivable system in the following
ways. First, the decisions can help to achieve a tradeoff
between the correct system operations and minimization of
service interruptions. Second, since the impact of damage to a
CIS tends to increase in time, the sooner the administrators
understands the survivability state of the system, the sooner
they can decide to repair the system, if necessary, to avoid
potential further losses. Hence, the dependability on the
system can be maintained.
Our Bayesian network based survivability decision network is shown in Figure 1. The network components will be discussed in detail in the next sections. We have also implemented the decision model using the Netica [17] tool, where each node represents a state, and a cause-effect relationship between two nodes is denoted by an arrow, called an edge. The testing cases that we conducted indicated this model is useful in various survivability related decisions.
A. Nodes
A node in a Bayesian network represents a variable in the
situation being modeled. Each node is represented graphically
by a labeled rectangle. There are three different kinds of
nodes in a Bayesian network: nature nodes, decision nodes
and utility nodes. A node can take different values, each of
which is referred to as a state of the node. In our survivability
decision model, there are nine nature nodes, two decision
nodes, and two utility nodes as shown in Figure 1. A nature
node models the nature or reality about one aspect of the
system. The semantic meanings of the nine nature nodes in
our model are described as follows.
1. The ―Security Vulnerability‖ node represents a security
property of a CIS system under consideration and
indicates how vulnerable the system is subject to
malicious attacks. From the system engineering
perspective, every system has a certain level of
vulnerability to known or unknown attacks. This node
has High, Medium or Low value to represent the
corresponding degree of vulnerability of a CIS.
Fig. 1. The Bayesian Network Based Survivability Decision Model
2. The node ―Adversary Capability‖ represents an
adversary’s ability to compromise a system. This node
has High, Medium or Low value to represent how
capable of an adversary to attack and compromise a CIS.
3. The combined effect of system security vulnerability and
adversary capability determines the likelihood of a
system being attacked. We use the node ―Hackable‖ to
represent how likely a CIS could be compromised given
the system’s vulnerability and the adversary’s ability to
attack.
4. System failures due to deficiencies of design and
implementation can endanger the survivability of the
system. The node ―System Fault‖ is used to represent
possible system failures. This node can have a High,
Medium or Low value, indicating the possibility of
system failures that could affect system survivability.
5. The effects of system failures are typically visible to
users in the form of degradation of services that the
system provides. The node ―System Performance‖
therefore represents the service quality provided by a CIS
as indicated by such system performance metrics as
network throughput, system response time, and service
quality. System performance is evaluated as normal or
poor.
6. A CIS system’s ability to conduct self-healing when
under attack or in a partial failure state is represented by
the node ―On-the-fly Repair‖. This ability is crucial to
automatically recover the system in a timely manner
before any more severe damage is resulted. This node can
have two values: ―Capable‖ or ―Not Capable‖, indicating
whether the system has the sufficient self-healing ability
from the system survivability perspective.
7. Similarly, fault tolerance allows a CIS to continue
operating properly in the event of failures. We represent
such a capability of the CIS using the ―Fault Tolerance‖
node in our model. The node has two possible values:
―Capable‖ or ―Not Capable‖, indicating whether the
system has sufficient ability to conduct fault tolerance for
the purpose of damage masking.
8. Given the values of the above 7 nodes, the condition of a
CIS is presented using a ―System Compromised or
Faulty‖ node. This node has two possible values: ―True‖
or ―False‖, which indicates whether the system is
compromised by adversary or in a faulty state due to
system failures.
Our survivability decision model has a provision for
system administrators to make a series of decisions regarding
system survivability. Each decision is represented by a
decision node (denoted as a rectangle with a different color
background from a nature node). Based on the system
condition represented by the ―System Compromised or
Faulty‖ node, system administrators can decide whether it is
necessary to perform thorough analysis of the system in order
to verify and confirm the survivability status of the system.
This decision is represented by the ―Do Offline Survivability
Diagnosis‖ decision node as shown in Figure 1. The diagnosis
result is represented by the ―Survivability Diagnosis‖ nature
node. Given the likelihood of system survivability status, the
system administrators can further decide whether to allow the
CIS to continuously support the mission critical services or it
is necessary take the system down for recovery. This decision
is represented by the ―Continue to Support Mission Critical
Services‖ decision node. For each decision node, the number
next to each choice indicates the expected utility of making
that choice. This expected utility is computed as a
multiplication of the utility of an event and the probability for
that event to occur, i.e., Expected Utility(A) = Utility(A) x
P(A). The highest value of the expected utility indicates the
best choice of that decision.
A utility node in a Bayesian network represents the utility
value of each choice associated with a decision. In our model
two utility nodes are specified: ―Diagnosis Cost‖ and
―Administration Action‖. The former represents the cost (a
non-positive utility value) associated with conducting an
offline survivability diagnosis. If the system administrators
decide that an offline survivability diagnosis is necessary, then
the cost is ―-10‖, which represent the time and efforts to check
the system as well as loss of productivity due to interruptions
of services to users. Otherwise, the cost is ―0‖ should no
diagnosis is required. The latter represents the utilities
corresponding to various decision choices regarding whether
the system should be allowed to continue to support mission-
critical services given its current status. For instance, if the
system is actually compromised or in a faulty state but still
allowed to provide critical services, then the consequence
would be unpredictable. Therefore, a negative utility value of
―-200‖ is assigned. In another case, if the system is not
compromised but the system administrators disallow it to
continue functioning (i.e., a false negative), then a utility value
of ―-50‖ is assigned in this case. This smaller negative utility
represents the cost of mistakenly interrupting the system when
the system is not compromised or in a fault state.
B. Edges
An edge of a Bayesian network represents a causality relationship between two nodes. It is represented graphically by an arrow between the two nodes and the direction of the arrow indicates the logical causality (see Figure 1). The intuitive meaning of an edge drawn from node A to node B is that A has a direct influence on B. The CPT associated with a node defines quantitatively how the node (called the child node) is influenced by others (called the parent nodes). In a relationship, the child node is conditionally dependent upon their parent nodes. For instance, in our model, the level of system security vulnerability and the capacity of adversary (the parent nodes) have great impact on the possibility of system being hackable (the child node). Therefore, there are causality relationships among those nodes. Furthermore, we identify the following factors that can affect a system’s security state (i.e., compromised or faulty): system on-the-fly repair ability, fault tolerance capacity, system hackability, and system possible internal failure (evidenced by system performance). In addition, the diagnosis results are determined by (1) the system administrator’s decision to take the offline system diagnosis; and (2) the hidden state of the system (compromised or faulty).
C. Conditional Probability Tables
In a Bayesian network, the relationship between two nodes
A and B is defined as a conditional probability, P(A | B),
which represents the probability of A conditional on a given
outcome of B. The conditional probability is calculated using
Bayes’ Theorem [18]:
P)(
)(*)|(
BP
APABP (1)
where P(B|A) is the conditional probability of B given A, and
P(B) and P(A) are the unconditional probabilities of B and A,
respectively. The probabilistic dependency is maintained by
the CPT, which is attached to the corresponding node. A CPT
shows all possible outcomes of a given node and the
conditional probability corresponding to each outcome given
all its parent nodes.
To present the conditional probabilities of parent nodes on
a child node, every intermediate and leaf node has a CPT
associated with it. For every possible combination of the
parent states, there is a row in the child node’s CPT that
describes the possible state that the child node should be.
Nodes with no parents also have CPTs but they only consist
of the probabilities for each state of that node. As an example
for an intermediate node, the CPT table in Figure 2 provides
an incident of causality relationship between the four parent
nodes: ―System Performance‖, ―Hackable‖, ―On-the-Fly
Repair‖, ―Fault Tolerance‖ and the child node ―System
Compromised or Faulty‖. The table lists all the possible
outcomes (and their possibilities) of the system’s security
status given the combination of the four influencing factors as
represented by the four parent nodes.
Fig. 2. CPT Table for the Node ―System Compromised or Faulty‖
Fig. 3. CPT Table Associated with the Node ―Survivability Diagnosis‖
Fig. 4. Bayesian Network Model Given the First Decision Choice to Conduct Offline Survivability Diagnosis
The values in the CPT table are added based on expert knowledge and they can be modified as more information is available. As another example, Figure 3 represents the possible diagnosis results if the system administrators decide to conduct a thorough system diagnosis for survivability verification. The values for the node ―Survivability Diagnosis' are influenced by its two parent nodes (see Figure 1): the nature node ―System Compromised and Faulty‖ and the decision node ―Do Thorough Offline Survivability Diagnosis‖. From the table, we can see that if the system is actually compromised or in a faulty state and the system administrators have decided to do a thorough system diagnosis, the test has 90% chance to be positive (i.e. system is compromised or Faulty) and 10% chance to be negative (i.e., the system is fault free). On the other hand, if the actual state of the system is fault free, the diagnosis result still have 40% chance to mistakenly report that the system is compromised or faulty. Those values are entered by users based on expert knowledge or they can be generated through a Bayesian learning process. Due to page limitation, we will not discuss those issues here.
D. Decision Scenarios
In a Bayesian network, a belief represents a user’s
judgment about the probability that a variable is in a certain
state based on available evidence. A piece of evidence is
information about a given fact about system properties.
Bayesian decision networks support vague, conflicting, and
incomplete evidence by allowing a user to enter a probability
for evidence of a variable being in each state. As shown in
Figure 1, given the evidences and the relationships among the
corresponding nodes, the expected utility of each decision
choice for ―Do Offline Survivability Diagnosis‖ is calculated
by following the Bayer’s Theorem [18]. As shown in Figure
1, the choice of not conducting offline survivability diagnosis
will result in a higher utility so it will be the best decision
given the current setting. Since the second decision, i.e.,
whether to allow the system to continue to function to support
mission-critical services, is dependent on the first decision,
i.e., whether to take the system offline for a full scale
diagnosis, no expected utilities are available in second
decision unless the first decision is made. If we assume that
TABLE I
TESTING CASES AND THE CORRESPONDING BEST DECISION CHOICES
the administrators decide not to do offline survivability
diagnosis (since this is the most rational decision with the
current evidence), then the expected utilities will be
automatically calculated for the node ―Continue to Support
Mission Critical Services‖ as shown in Figure 4. From the
values of this second decision node, the best choice is to allow
the CIS system to continue supporting mission critical services
since this decision will result in a higher utility.
In addition to the above testing case using the default
values, we have conducted various other testing with different
user input as additional evidence for system states. Those
testing help answer various ―what-if‖ questions regarding
different decision scenarios given different system status
already known. The testing cases we conducted and the
corresponding best decision choices are summarized in Table
I. The highlighted columns (i.e., DOSD and CTSMS)
represent the two sequential decisions that a system
administrator needs to make.
Each entry in Table I represents a global system state with
a set of probabilistic values for each node. An entry is
composed of multiple cells. While a single-value cell
represents user-provided evidence (known fact about a system
property), a multi-value cell represents a set of possible states
of the node with the calculated probability of occurrence. We
use an example to explain a testing case. Consider the fourth
entry, where the following information is given as user input
evidence and the corresponding best decisions are provided:
1. System vulnerability is low (as indicated by the value
―Low‖ for SV);
2. Adversary ability to compromise system is low (as
indicated by the value ―Low‖ for AC);
3. System is capable of performing on-the-fly repair (as
indicated by the value ―Capable‖ for OFR);
4. System is capable of providing fault tolerance in event
of failures (as indicated by the value ―Capable‖ for
FT);
5. System fault is low (as indicated by the value ―Low‖
for SF);
6. System performance is normal (as indicated by the
value ―Normal‖ = 100% for SP); 7. Likelihood of system being hackable is low (as indicated
by the value ―Low‖=100% for HA).
With the above information, the possibility of ―System Compromised or Faulty‖ is updated by the Bayesian network as ―True‖ = 10% and ―False‖ = 90%. That means that the system has 90% chance of not being compromised or in a faulty state. Then, based on this information, the system administrators should best choose not to do offline survivability diagnosis because this choice will result in the highest utility (i.e., 8.0 as shown for DOSD). Consequently, the administrator should decide to allow the system to continue supporting mission critical services since it is less likely that the system has been compromised or in a faulty state. Other testing case (e.g., entry 3) may suggest the system administrators to conduct solid system diagnosis. If the diagnosis results are negative, the system should be suspended to schedule a repair.
IV. CONCLUSION
Survivability related decisions are dependent on various
factors and can be divided into a series of stages. The model
proposed in the paper is developed based on Bayesian network
which is a powerful tool to represent cause-effect and
probabilistic relationships. We identify a set of dependent
factors that affect the survivability of a system and then define
the probabilistic relationships among those factors and the
system states. The model helps system administrators to make
Entry SV AC OFR FT SF SP (%) HA (%) SCF
(%)
DOSD SDR (%) CTSMS
1 High High Not
Capable
Not
Capable
High Normal = 20
Poor = 80
High = 65
Medium = 30
Low = 5
True =
81.5
False =
18.5
No =
-41.68
None =100
Positive = 0
Negative = 0
No =
-41.675
2 High High Not
Capable
Not
Capable
Low Normal = 80
Poor = 20
High = 65
Medium = 30
Low = 5
True =
75.5
False = 24.5
No =
-38.975
None =100
Positive = 0
Negative = 0
No =
-38.975
3 High High Not
Capable
Not
Capable
Low Normal = 80
Poor = 20
High = 65
Medium = 30
Low = 5
True =
75.5
False =
24.5
Yes =
-70.60
None =100
Positive = 0
Negative = 0
Yes =
-70.60
4 Low Low Capable Capable Low Norma l
=100
High = 0
Medium = 0
Low= 100
True =
10
False =
90
No
=8.0
None =100
Positive = 0
Negative = 0
Yes =
8.0
5 Low High Not
Capable
Not
capa
ble
Low Norma l =
100
High = 30
Medium = 20
Low= 50
True =
57
False = 43
No =
-30.65
None =100
Positive = 0
Negative = 0
No =
-30.65
a good judgment on the current state of the system given
various evidences. It also helps users determine (1) whether it
is necessary to conduct a solid system diagnosis for potential
survivability problems of the system; and (2) based on the
diagnosis results (if system testing is conducted), whether the
system should be allowed to continue function or it should be
suspended for repair to avoid further damage and future losses.
The Bayesian networks are adaptable and can perform well
with limited information. While our model only considers the
most influential factors on the survivability of a system, it can
be enhanced when more information is available when applied
to specific applications.
ACKNOWLEDGMENT
The authors are thankful to Dr. Robert Herklotz for his support, which made this work possible.
REFERENCES
[1] Knight, J., M. Elder, J. Flinn, and P. Marx. ―Summaries of Four Critical Infrastructure Systems,‖ Technical Report CS-97-27, Department of Computer Science, University of Virginia, November 1997.
[2] Elder, M. C. ―Fault Tolerance in Critical Information Systems‖, Doctoral Thesis. UMI Order Number: AAI3012667, University of Virginia, 2001.
[3] ―Critical Foundations: Protecting America’s Infrastructures‖, The Report of the President’s Commission on Critical Infrastructure Protection, United States Government Printing Office (GPO), No. 040-000-00699-1, October 1997.
[4] Ellison, R., Fisher, D., Linger, R., Lipson, H., Longstaff, T. and Mead. N. ―Survivable Network Systems: An Emerging Discipline,‖ Technical Report CMU/SEI-97-TR-013, Software Engineering Institute, Carnegie Mellon University, November 1997.
[5] Hansson, Sven Ove, "Decision Theory: A Brief Introduction", retrieved from http://www.infra.kth.se/~soh/decisiontheory.pdf on January 12, 2010.
[6] Knight, J., Sullivan, K., Elder, M. and Wang, C. ―Survivability architectures: issues and approaches‖, DARPA Information Survivability Conference and Exposition, pp. 157-17, 2000.
[7] Robert J. Ellison, David A. Fisher, Richard C. Linger, Howard F. Lipson, Thomas A. Longstaff, Nancy R. Mead, "Survivability: Protecting Your Critical Systems," IEEE Internet Computing, vol. 3, no. 6, pp. 55-63, Nov./Dec. 1999, doi:10.1109/4236.807008
[8] Zhu, J. and Wang, C. "A Survivable Scheme for Critical Information System," 2008 International Conference on Security Technology, pp.230-235, 2008.
[9] Sullivan, K., Knight, J. C., Du, X., and Geist, S. ―Information survivability control systems‖, The 21st International Conference on Software Engineering, pp. 184-192, Los Angeles, California, United States, May 16 - 22, 1999.
[10] Modelo-Howard, G., Bagchi, S., and Lebanon, G. ―Determining Placement of Intrusion Detectors for a Distributed Application through Bayesian Network Modeling‖, The 11th international Symposium on Recent Advances in intrusion Detection, Cambridge, MA, USA, September 15 - 17, 2008.
[11] Cemerlic, A., Yang, L. and Kizza, J. ―Network Intrusion Detection Based on Bayesian Networks‖, Proceedings of Software Engineering and Knowledge Engineering, July 2008.
[12] Frigault, M., Wang, L., Singhal, A. and Jajodia, S. ―Measuring network security using Dynamic Bayesian network‖, Proceedings of 4th ACM Workshop on Quality of Protection (QoP), October 27, 2008, p. 23-30.
[13] Bringas, P.,Penya, Y., Paraboschi, S. and Salvaneschi, P. ―Bayesian-Networks-Based Misuse and Anomaly Prevention System‖, Proceedings of the Tenth International Conference on Enterprise Information Systems, p. 62-69, Barcelona, Spain, June 12-16, 2008.
[14] Abouzakhar, N., Gani, A., Manson, G., Abuitbel, M. and King, D., ―Bayesian Learning Networks Approach to Cybercrime Detection‖, Proceedings of the 2003 PostGraduate Networking Conference, Liverpool, United Kingdom, 2003.
[15] Mukhopadhyay, A., Chatterjee, S., Saha, D., Mahanti, A. and Sadhukhan, S. " e-Risk Management with Insurance: A framework using Copula aided Bayesian Belief Networks," Proceedings of the 39th Hawaii International Conference on System Sciences, Hawaii, USA, January, 2006.
[16] Herath, H. and Herath, T. ―Investments in Information Security: A Real Options Perspective with Bayesian Postaudit‖, Journal of Management Information Systems, Vol.25, No.3, p.337-375, September, 2008.
[17] Home Page of Norsys Software Corporation, http://www.norsys.com.
[18] Jensen. F. ―Bayesian Networks and Decision Graphs‖, Springer–Verlag New York, 2001.
Keynote: So You Wanna Be a Bot Master?
Billy Rios Security Researcher, Google, Inc.
OLLOW me as we explore the command and control software used by Bot Masters, analyze the tools
used by the shadiest characters in the underground, and uncover data being stolen by bots from their
unsuspecting victims. We'll dive into the source code and dissect the individual components of a real
botnet command and control server. We'll also take a detailed look at what types of data the bots are after
and how the bot gets the stolen data back to its master. All the software is real and all the data is genuine.
You'll get to see the actual control consoles used by Bot Masters as they control an army of bots and learn
to find these servers on the Internet.
F
Abstract—The popularity of peer-to-peer (P2P) file distribution is
consistently increasing since the late 1990’s. In 2008, P2P traffic
accounted for over half of the world’s Internet traffic. P2P
networks lend themselves well to the unauthorised distribution of
copyrighted material due to their ease of use, the abundance of
material available and the apparent anonymity awarded to the
downloaders. This paper presents the results of an investigation
conducted on the top 100 most popular BitTorrent swarms over
the course of one week. The purpose of this investigation is to
quantify the scale of unauthorised distribution of copyrighted
material through the use of the BitTorrent protocol. Each IP
address, which was discovered over the period of the weeklong
investigation, is mapped through the use of a geolocation
database, which results in the ability to determine where the
participation in these swarms is prominent worldwide.
I. INTRODUCTION
N 2008, Cisco estimated that P2P file sharing accounted for
3,345 petabytes of global Internet traffic, or 55.6% of the
total usage. Cisco forecast that P2P traffic would account for
9,629 petabytes globally in 2013 (approx. 30% of the total
usage) [3]. While the volume of P2P traffic is set to almost
triple from 2008-2013, its proportion of total Internet traffic is
set to decrease due to the rising popularity of content
streaming sites and one-click file hosting sites such as
Rapidshare, Megaupload, etc. BitTorrent is the most popular
P2P protocol used worldwide and accounts for the biggest
proportion of Internet traffic when compared to other P2P
protocols. Depending on region, Ipoque GmbH has measured
BitTorrent traffic to account for anything from 20%-57% of
that region’s total Internet usage in 2009 [11].
The BitTorrent protocol is designed to easily facilitate the
distribution of files to a very large number of downloaders
with minimal load on the original file source [2]. This is
achieved through the downloaders uploading their completed
parts of the entire file to other downloaders. A BitTorrent
swarm is made up of seeders, i.e., peers with complete copies
of the content shared in the swarm, and leechers, i.e., peers
who are downloading the content. Due to BitTorrent’s ease of
use and minimal bandwidth requirements, it lends itself as an
ideal platform for the unauthorized distribution of copyrighted
Co-funded by the Irish Research Council for Science, Engineering and
Technology and Intel Ireland Ltd. through the Enterprise Partnership Scheme.
material, which typically commences with a single source
sharing large sized files to many downloaders.
To commence the download of the content in a particular
BitTorrent swarm, a metadata “.torrent” file must be down-
loaded from an indexing website. This file is then opened
using a BitTorrent client, which proceeds to connect to several
members of the swarm and download the content. Each
BitTorrent swarm is built around a particular piece of content
which is determined through a unique identifier based on a
SHA-1 hash of the file information contained in this UTF- 8
encoded metadata file, e.g., name, piece length, piece hash
values, length and path.
Each BitTorrent client must be able to identify a list of
active peers in the same swarm who have at least one piece of
the content and is willing to share it, i.e., that has an available
open connection and has the bandwidth available to upload.
By the nature of the implementation of the protocol, any peer
that wishes to partake in a swarm must be able to
communicate and share files with other active peers. There are
a number of methods that a client can attempt to discover new
peers who are in the swarm:
1. Tracker Communication – BitTorrent trackers
maintain a list of seeders and leechers for each
BitTorrent swarm they are currently tracking. Each
BitTorrent client will contact the tracker
intermittently throughout the down- load of a
particular piece of content to report that they are
still alive on the network and to download a short
list of new peers on the network.
2. Peer Exchange (PEX) – Peer Exchange is a
BitTorrent Enhancement Proposal (BEP) whereby
when two peers are communicating, a subset of
their respective peer lists are shared during the
communication.
3. Distributed Hash Tables (DHT) – Within the
confounds of the standard BitTorrent specification,
there is no intercommunication between peers of
different BitTorrent swarms. Azureus/Vuze and
μTorrent contain mutually exclusive
implementations of distributed hash tables as part
of the standard client features. These DHTs
maintain a list of each active peer using the
corresponding clients and enables cross-swarm
A Week in the Life of the
Most Popular BitTorrent Swarms
Mark Scanlon, Alan Hannaway and Mohand-Tahar Kechadi 1
UCD Centre for Cybercrime Investigation, School of Computer Science & Informatics,
University College Dublin, Belfield, Dublin 4, Ireland
{mark.scanlon, alan.hannaway, tahar.kechadi}@ucd.ie
I
communication between peers. Each peer in the
DHT is associated with the swarm(s) in which he
is currently an active participant.
The most popular BitTorrent indexing website, according to
Alexa, is The Pirate Bay [9]. In January 2010, The Pirate Bay
held the Alexa global traffic rank of number 99 and is the 91st
most popular website visited by Internet users in the United
States [1]. For the purpose of the investigation outlined in this
paper, the top 100 torrents listed on The Pirate Bay were
chosen to be investigated due to the popularity of the website.
II. SPECIFICATIONS OF THE INVESTIGATION
The steps involved in the execution of this investigation are:
1. Connect to The Pirate Bay and download the
“.torrent” metadata files for the top 100 torrents.
2. Connect to each swarm sequentially and identify each
of the IP addresses currently active in the swarm until
no new IPs are found.
3. Once the active IPs are found for the entire 100
torrent swarms, the process is repeated for the next
24 hours.
4. After 24 hours, the process was begun again at step 1.
The investigation was conducted using a single dedicated
server, which sequentially monitored each torrent swarm until
all the IPs in the swarm were found. Over the course of the
seven day investigation, a total of 163 different torrents were
investigated. None of the content appearing in these torrents
was found to be distributed legally; each torrent swarm is
distributing copyrighted material without any documented
authorization.
A. Investigation Methodology
For a regular BitTorrent user, the starting point to acquiring
a given piece of content is to visit a BitTorrent indexing
website, e.g., The Pirate Bay [9]. The indexing site serves as a
directory of all the content available to the user. The user finds
the specific content of interest by browsing the website or
through keyword search. He must then download the
“.torrent” metadata file which contains information specific to
the content desired such as name, size, filenames, tracker
information, hash value for completed download, chunk size,
hash values for each chunk, etc.
This “.torrent” file is then opened with a BitTorrent client,
e.g., Azureus/Vuze, μTorrent, etc., which in turn contacts the
tracker or the DHT for a bootstrapping list of IP addresses to
get started in the swarm.
The main goal of the methodology used for this
investigation is to gather information in an intelligent and
novel manner through the amplification of regular client
usage. This is in order to collect the complete list of IPs
involved in any given swarm as efficiently as possible. For
example, in a large swarm of >90,000 IPs, the software tool
developed for this experiment is capable of collecting the
active peer information in as little as 8 seconds. However, the
precise design and specifications of the software used during
the investigation outlined is beyond the scope of this paper.
B. Specifics of the Content Investigated
From the analysis of the daily top 100 swarms, video
content was found to be the most popular being distributed
over BitTorrent. Movie and television content amounted for
over 94.5% of the total, as can be seen in Fig. 1, while music,
games and software amounted for 1.8%, 2.5% and 1.2%
respectively. One highly probable explanation for the
popularity of television content is due to the lag between US
television shows airing in the US and the rest of the world.
Fig. 1. BitTorrent swarms investigated by category
III. RESULTS AND ANALYSIS
Over the week long investigation, the total number of unique
IP addresses discovered was 8,489,287. On average, each IP
address detected was active in 1.75 swarms, or, almost three
out of every four IP addresses were active in at least two of the
top 100 swarms during the week. The largest swarm detected
peaked at 93,963 active peers and the smallest swarm shrunk
to just 1,102 peers. The time taken to capture a snapshot of
100 swarms investigated varies due to the increase and
decrease in the overall size of the swarms. The average time to
collect all the peers’ information for each swarm is 3.4
seconds.
A. File Information
Notably, 50.6% of the files contained in the top 100 torrents
are split into smaller chunks for distribution, as can be seen in
Figure 2. 38.9% of the files were also compressed into RAR
files to save on the overall size of the content. The partitioning
TABLE I BREAKDOWN OF THE INFORMATION CONTAINED
IN THE “.TORRENT” METADATA FILES
Maximum Minimum Average
Content Size 38.37GB 109.69MB 1.62GB
Chunk Size 4MB 128KB 1.3MB Number of Chunks 19,645 346 1,251
Number of Files 322 2 24
Number of Trackers 57 4 20
of large files into numerous smaller files is consistent with file
distribution through one-click file hosting websites and
newsgroups where the distribution of small files greatly
improves the overall throughput of the system.
Fig. 2. Breakdown of file types discovered during the investigation
Video files are distributed as AVI, MP4 or MKV files and
are typically grouped with screenshots, subtitles and sample
videos. Music files are generally distributed as MP3 files and
are grouped with associated album playlists and artwork.
B. Worldwide Distribution
Fig. 1. Top 10 countries detected.
There are a number of assumptions that had to be made for
the purposes of the geolocation of the IP addresses and for the
count of the total number of BitTorrent users involved in the
swarms investigated:
1. The country, city and ISP level geolocation
databases used are as accurate as possible. In
January 2010, MaxMind state that their
geolocation databases are 99.5% accurate at a
country level, 83% accurate at a US city level
within a 25 mile radius and 100% accurate at the
ISP level [7].
2. For a week long investigation, each IP address
found to be participating in the swarms
investigated is assumed to only ever be allocated to
one end user. Due to a typical DHCP lease from an
Internet service provider lasting somewhere in the
order of 2-7 days, dynamic IP address allocation
may result in the reallocation of the same IP
address to two or more end users during the
investigation. Should this have occurred during the
investigation, it is ignored for the interpretation of
the results outlined below. It is deemed technically
infeasible to identify precisely when this may
occur on a global level within the scope of this
paper.
3. No anonymous proxy servers or Internet
anonymity services, e.g., I2P [5], Tor [12], etc., are
used by the IP addresses discovered.
4. It is infeasible for users on dial-up Internet
connections to download the very large file that
typically justifies distribution via the BitTorrent
protocol. The average content size for the swarms
investigated was 1.62GB, which would take a
typical 56kbps dial-up user over 69.5 hours to
download, assuming no other Internet traffic and
achieving the maximum theoretical dial-up
connection speed. For the purposes of the analysis
presented in section III below, it was assumed that
a negligible amount of dial-up users participate in
the BitTorrent swarms.
For each IP address detected during the investigation, the
geolocation is obtained using MaxMind’s GeoIP databases
[7], which results in information such as city, country, latitude,
longitude, ISP, etc., being resolved. This information is then
gathered and plotted as a heatmap to display the distribution of
the peers involved in copyright infringement on a world map,
seen in Figure 4. The most popular content tends to be content
produced for the English speaking population, which is
reflected in the heatmap, i.e., countries with a high proportion
of English speaking population are highlighted in the results.
TABLE II
NUMBER OF BROADBAND SUBSCRIBERS DISCOVERED DURING THE
INVESTIGATION (ASSUMING A NEGLIGIBLE AMOUNT OF DIAL-UP USERS)
Country
Number of
IP Addresses
Discovered
Broadband
Subscription
Count [6]
Percentage of
Broadband Subscriptions
Discovered
United States 1,116,111 73,123,400 1.53% United Kingdom 790,162 17,276,000 4.57%
India 549,514 5,280,000 8.70%
Canada 443,577 9,842,300 4.51% Spain 397,892 8,995,400 4.42%
Italy 378,892 8,995,400 3.36%
Brazil 320,829 10,098,000 3.18% Australia 261,433 5,140,000 5.09%
Poland 248,731 4,792,600 5.19%
Romania 215,403 2,506,000 15.4%
A significant percentage of the worldwide broadband sub-
scribers were detected during the investigation. 2.43% of the
349,980,000 worldwide broadband subscriptions was
discovered during the investigation [6]. The percentages of
broadband subscribers detected in the top 10 countries are
outlined in Table II. The top ten countries detected account for
over 53.6% of the total number of IPs found.
C. United States
The United States is the most popular country detected with
over 1.1 million unique IP addresses, which accounted for
13.15% of all the IP addresses found. While accounting for the
largest portion of the results obtained in this investigation, this
relatively low percentage suggests that BitTorrent has a much
more globally dispersed user base in comparison to other large
P2P networks. For example, a 10 day investigation conducted
on the Gnutella network in 2009, it was found that “56.19% of
all [worldwide] respondents to queries for content that is
copyright protected came from the United States” [4]. When
the IP addresses detected during this investigation are
geolocated and graphed onto a map, the population centers can
be easily identified, as can be seen in Figure 5. The state of
California accounted for 13.7% of the US IPs found, with the
states of Florida and New York accounting for 7.2% and 6.8%
respectively.
D. Extrapolated Results
If the total amount of worldwide public BitTorrent activity is
considered to be the summation of the number of active IP
addresses in all BitTorrent swarms at any given moment
served by the world’s largest trackers, the percentage of
overall BitTorrent activity analyzed as part of this
investigation can be estimated. From analyzing the scrape
information available from two of the largest trackers,
OpenBitTorrent [8] and PublicBitTorrent [10], it is estimated
that the most popular 100 torrents at any given moment
accounts for approximately 3.62% of the total activity.
IV. CONCLUSION AND FUTURE WORK
The objective of this investigation was to attempt to identify
the scale of the unauthorized distribution of copyrighted
material worldwide using the BitTorrent protocol. 2.43% of
the broadband subscriber base was detected over the course of
one week in the 163 torrents monitored. This number is far
greater than the number of subscribers that could possibly be
prosecuted for their actions. The number of end users involved
in illegal downloading is undoubtedly much higher than this
due to the relatively small scale of this investigation. Some
network factors will also have a negative effect over the
results achieved, such as two or more end-users appearing as a
Fig. 2. Heatmap Showing the Worldwide Distribution of Peers Discovered.
single Internet IP address though Internet connection sharing,
proxy services etc.
To further improve the results outlined above, a number of
additional steps would be considered:
1. Overcome dynamic IP address reallocation – While
a number of the IPs discovered during the
investigation may have been allocated to more than
one end user, this multiple allocation is not possible
to identify using the standard BitTorrent protocol.
However, the identification of individual users
through metadata examination and heuristic
approaches is possible. This would increase the
overall accuracy of the results obtained and give an
accurate measure of the total BitTorrent population.
2. Conduct a larger scale analysis – In each 24 hour
period of this investigation, only 100 swarms were
monitored. Increasing the scale of the investigation
will yield to better results. The scaled-up analysis
of all BitTorrent swarms will identify some
interesting statistics, e.g., legal vs. illegal
distribution, total worldwide BitTorrent population,
etc.
3. Identification of peers using anonymous Internet
services – By comparing the IP addresses
discovered during the investigation with a list of
known Internet traffic proxy or pass-through
services, such as that maintained by MaxMind [7],
the quality of the results collected can be greatly
improved.
4. Identification of Internet connection sharing –
Through the analysis of the peer metadata, such as
client information, downloaded content types and
categories, etc., multiple users sharing a single
Internet connection can be identified.
REFERENCES
[1] Alexa Information on The Pirate Bay, Downloaded January 2010,
http://www.alexa.com/siteinfo/thepiratebay.org
[2] The BitTorrent Protocol Specification, Downloaded January 2010,
http://www.bittorrent.org/beps/bep 0003.html
[3] Cisco Systems, Inc., Cisco Visual Networking Index: Forecast and
Methodology, 2008-2013. Retrieved from: http://www.cisco.com/en/
US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white paper
c11- 481360.pdf. 2009.
[4] A. Hannaway and M-T. Kechadi, An Analysis of the Scale and Distribu-
tion of Copyrighted Material On The Gnutella Network International
Conference on Information Security and Privacy, Orlando, USA, 2009.
[5] I2P - The Anonymous Network, http://www.i2p2.de
[6] International Telecommunication Union, United Nations, Report on
Internet, Report Generated January 2010, http://www.itu.int
[7] MaxMind Inc., GeoLite Country Database, Downloaded January 2010,
http://www.maxmind.com
[8] OpenBitTorrent, Aggregated Scrape File. Downloaded January 2010,
http://openbittorrent.com/all.txt.bz2
[9] The Pirate Bay, World’s Largest BitTorrent Tracker, Total Top 100,
Downloaded January 2010, http://thepiratebay.org/top/all
[10] PublicBitTorrent, Aggregated Scrape File. Downloaded January 2010,
http://www.publicbt.com/all.txt.bz2
[11] H. Schulze, K. Mochalski, Internet Study 2008/2009. Ipoque GmbH.
http://www.ipoque.com/resources/internet-studies/ 2009.
[12] The Tor Project, http://www.torproject.org/
Fig. 5. Distribution of IP addresses in mainland United States. The largest circles represent locations with
greater than 1500 IPs, while the smallest circles may only represent a single IP at a given location
Abstract—Universities consider lectures and classes to be
intellectual property which must be protected. In this case
study we discuss the processes by which outreach courses
delivered via the Internet are protected. Specifically, we
describe two procedures used to protect outreach courses at
our University. In this setting, the School of Pharmacy uses a
streaming media system provided by Starbak system, while
the College of Engineering and College of Business utilize a
streaming system developed by Sony. Additionally, we
present a comparison of these two methods from an
Information Assurance standpoint. We also present the
different domains of CISSP (Certified Information Systems
and Security Professional) as applicable to video streaming
and discuss how the security triad concepts of confidentiality,
integrity, and availability are addressed by the differing
platforms.
Index Terms—Access Control, Role-based Access Control,
Physical Security
I. INTRODUCTION
RIOR to 1995 integrating audio or video into
websites was a major problem. Before playing a video
or an audio file it had to be downloaded to the local hard
drive and, depending on the file size and network
connection (which was usually dial up), it would take
anywhere from five minutes to two hours to download.
With the advent of new technology, however, this changed
and now with the help of streaming technology one can
listen to or watch media files as they are downloaded.
Because of the difficulties with distributing classes over
the Internet, most of the universities that engaged in
distance education at that time chose to distribute movies,
either as VHS or DVD recordings, to the students.
The rise of streaming technology enabled universities to
provide outreach and distance education courses far more
efficiently than ever before. The use of streaming
technology also introduced several new difficulties for
administrators interested in protecting the University’s
intellectual property. Whereas someone was previously
required to have physical access of some kind, either by
accessing the recording equipment or by intercepting the
physical media that was mailed to the students, the advent
of streaming media meant that thieves could electronically
steal intellectual property at several different places in the
distribution process.
In this case study, we discuss the methods by which
classes are streamed, stored and transmitted to students
located at different geographical locations. Moreover, we
will examine how the different domains of the CISSP
certification are satisfied in order to protect the
intellectual property of the University. We will look into
aspects such as Access Control, Business Continuity/
Disaster Recovery, Legal and Regulatory Compliance and
Investigation, Operations Security and Physical Security
(though not necessarily in that order). We will consider
two different platforms used for streaming classes at our
University: the Starbak system, used by the School of
Pharmacy [1] and a system developed by Sony, used by
the College of Engineering and the College of Business.
The rest of this paper is organized as follows. Section
two presents a brief overview of how classes are recorded
and distributed using each system. Section three examines
the security measures undertaken on both the Sony and
Starbak systems. Section four describes our future work
and in section five we offer some conclusions.
II. OVERVIEW OF CLASSROOM RECORDING AND
STREAMING
A. Distributing Courses with Sony Equipment
Most departments at our University record and store the
classes that are conducted for on-campus students. These
courses are also made available to students that have
enrolled in the distance education program. On occasion
the courses are streamed live for distance education
students, though this is not the norm. The College of
Engineering and College of Business use identical
equipment and procedures to stream classes to distance
education students, so we shall consider them together.
Sriharsha Banavara, Eric Imsand, and John A. Hamilton, Jr. Department of Computer Science & Software Engineering
Auburn University Auburn, AL 36849
{sbs006, eimsand, hamilton}@auburn.edu
Information Assurance in Video Streaming:
A Case Study
P
The classrooms in which the lectures are recorded are
fitted with three or four robotic Sony HD cameras. A high
quality microphone is used to record the lecturer’s voice.
The equipment operator starts recording by use of a GUI
application specially designed and built by the University
Engineering department. This GUI has facility to start,
stop, pause the video and is integrated to both the
streaming server and backup server. The basic resolution
of the video is 1024 by 768. This system uses the
Windows Media Encoder. Once the operator stops the
video recording the recorded video is converted into three
different formats: MP3, MP4, and Windows Media. After
conversion the files are temporarily stored on the local
computer. This is done with the help of automated Visual
Basic scripts that are scheduled to run as soon as the
operator closes the GUI window. They are pushed onto
the server as soon as the conversion is complete. In case
this fails for some reason, there is another script that
attempts to push any local files onto the main server once
every half an hour. The students can then login to the web
portal and access the videos in any format. They even
have the option of downloading them and viewing at a
later point in time.
Total number of classes that were streamed in the year
2009 by the College of Engineering and College of
Business put together is approximately 5000. The duration
for each class period varies between 45 minutes to an hour
and fifteen minutes. The amount of space required for an
hour of video recording is around 300MB. This adds up to
approximately 2 terabytes per year of needed storage.
B. Distributing Courses with the Starbak System
There are several differences in the procedure followed
by the School of Pharmacy and the one used by the
Colleges of Business and Engineering. The primary
reason for these differences is the intended audience. The
Colleges of Business and Engineering support a
traditional distance education program: individual
students located at geographically dispersed locations.
The School of Pharmacy, on the other hand, provides
lecture footage to large groups of students all physically
located in the same place taking the course at a satellite
campus.
The other key difference between the School of
Pharmacy and the Colleges of Business and Engineering
is the need for two-way communication. Students at the
satellite campus need to be able to engage in
videoconferencing so that they can ask the instructor
questions just as any other student would.
A high quality microphone is used to record the
lecturer’s voice. Each classroom where videos are
recorded is equipped with Polycom RSS 2000, which has
a built-in camera for the purpose of recording video in
real time. Polycom RSS 2000 is an on-demand recording,
streaming, and archiving solution for multimedia
conferences. RSS 2000 recording and streaming server
provides high capacity internal storage to archive
recorded conferences to meet legal compliance. It is easy
to record using the RSS 2000 from any type of video
conferencing endpoint as it uses just the basic commands
such as Start, Pause and Stop [2]. Around fifteen to
twenty classes are streamed weekly and account for
approximately fifty hours per week on an average, which
means nearly two hundred hours per month. The RSS
2000 system is capable of handling up to nine hundred
hours of streaming per month. In the year 2009, more than
1,100 classes were streamed with the duration of each
class varying between 1 to 4 hours. More recently RSS
4000 has been launched that has more features and is
supposed to be more robust than its predecessor.
The Starbak system consists of a conference engine,
encoder, delivery node and a Manager portal as shown in
Figure 1. Video is captured using Polycom RSS 2000 and
the two streams of data are sent to the Conference Engine.
Conference Engine constantly interacts with the Encoder.
Encoder produces .wmv and .asf extension files that are
part of the Windows Media framework. The Starbak
system is capable of handling ten simultaneous dual
screen recordings since it has ten encoders. Once that is
done, Delivery node takes the responsibility of forwarding
it to the Manager portal. It is here that the students and
faculty members can login with username and password to
access the videos. Username and password is the same for
all students of SOP. Native resolution supported by the
system is 352 by 288 pixels. The videos are available on
the Windows media player that is embedded on the VSM
portal. One can also change the quality of the video
Fig. 1. Conceptual layout of the Starbak System
depending upon the internet speed. VSM automatically
adjusts the resolution depending upon speed of the
internet connection. For example, with actual download
speed of 375kbps it adjusts to 640 by 480 resolution.
III. APPLICATION OF CISSP DOMAINS AND PRINCIPLES TO
STREAMING COURSES
The advent of the Internet has given rise to many bad
things: piracy, data theft, intellectual property theft, etc.
From top secret defense documents to educational
institutions, no content on the internet can be claimed to
be 100% safe. It can be argued that educational
institutions face attack from an unusually large population
demonstrating an extremely large range in skill and
ability. Videotaped lectures are at risk of falling into the
wrong hands and being hosted on the Internet for free or
being distributed to another University. Thus there is
tremendous concern regarding plagiarism and IP theft.
Ensuring that adequate attention is given to the way data
is processed, stored and distributed is critical. As a result,
Information Assurance is at the top of most people’s list
of concerns. Therefore we shall examine confidentiality,
integrity and availability with regard to the distribution of
streamed lectures.
Like any good science, information assurance must be
undertaken with a high degree of discipline. For this
reason we will frame our discussion against the domains
of the common body of knowledge (CBK) of the CISSP
examination. The CISSP exam is held in very high regard
amongst information assurance professionals. People
holding the CISSP certification are among the highest
paid in the field. There are 10 domains in the CISSP
CBK. We will focus on five of those ten domains, each
listed under separate headings.
A. Access Control
Access controls can be defined as mechanisms that
protect the assets of an enterprise. They reduce exposure
to unauthorized activities and allow access to information
systems only to authorized personnel. In other words, they
specify which users can access the system, what resources
they can access, the operations they can perform and also
provide individual accountability [4].
In this case, the asset to be protected is the intellectual
property – the class lecture and video tape. Access
control is the process of ensuring only authorized users
are able to access the class lectures. Furthermore, the
system must ensure that legitimate users only have access
to the features and content to which they are entitled.
There are two kinds of access control. Role based
access control and rule based access control. Of the two,
Role Based Access Control is used to regulate access to
the protected material. Students log into the system using
their University generated username and password.
Videos can be viewed by both the distance education
students as well as on-campus students. Students and
faculty members login to the same portal to view the
videos, but encounter different views. In the student view,
users can view videos for the classes in which they are
enrolled. Faculty also have this capability as well as the
ability to request that the administrator lock certain videos
to certain students. An example of this occurs when a
faculty member wants to discuss the results of an exam
that a student or students have not yet taken (possibly due
to illness or other approved absence). This approach is the
same for all three colleges at the University.
B. Physical Security
Physical security is the portion of the Common Body of
Knowledge that addresses the common physical and
procedural risks that exist in the environment in which
physical systems are protected [4].
The College of Business, College of Engineering, and
School of Pharmacy server rooms all have considerable
physical security. The rooms are equipped with locks and
security camera’s fixed on the outside walls. In order to
enter the room one is required to swipe in their ID card
and type in a unique pin number. Thus they have multi-
factor authentication in place. The cases in which the
servers are present have a special combination lock so that
they cannot be detached easily.
At the SoP, the maintenance for these servers is done
by the Starbak Company as part of the Service Level
Agreement. As a result, only two University personnel
are issued keys for the room: the IT Specialist who
controls the streaming system and his/her supervisor. A
copy is also present in the Dean’s Office. If an intruder
manages to gain access to the server room and tamper
with the server in any way, tamper sensors will notify the
company which is monitoring the server. They in-turn
will contact the SoP to verify the alarm. If this activity
was not part of an authorized activity, the call serves to let
them know that there has been a breach in security.
C. Business Continuity and Disaster Recovery
Business Continuity and Disaster Recovery efforts are
designed to ensure that normal operations will continue in
the event of a major event, such as a natural disaster [4].
With regard to streaming media for coursework, the
primary business activity is the serving of video. A
secondary concern is the preservation of data. To this
end, the colleges store backup copies of all videos in three
physically separate locations. There is a backup on the
local system, one copy on server and another on DVD.
Furthermore, in the case of the School of Pharmacy,
backup copies are also stored at the satellite campus,
providing a truly offsite backup solution. The Colleges of
Business and Engineering choose to store their backups in
different buildings on-campus, thereby mitigating the
threats posed by fire, flood, and wind-storm. These
preparations increase the likelihood that rapid recovery
from a disaster will be possible. Finally, once per year,
copies of all videos are delivered to the professors giving
each class as an additional form of safe archival.
D. Telecommunications and Network Security
The domain known as Telecommunications and
Network Security encompasses transmission methods,
formats and security measures used to provide protection
of the material while it is in transit over private and public
communications network and media [4].
The University uses firewalls and intrusion detection
systems to thwart head-on assaults by attackers. The
University has deployed modern, up-to-date anti-virus
software on all computer systems joined to the network in
order to slow and/or halt the spread of network based
worms as well as to detect malware which could offer
attackers back-door access. The University uses
encrypted protocols like Secure HTTP (https) to prevent
the sniffing of credentials by attackers.
E. Legal Compliance
According to the Common Body of Knowledge, this
domain addresses computer crime legislation and
regulations, investigative measures and techniques that
used to determine if an incident has occurred [4]. At the
orientation for distance education students, as well as for
on-campus students, the “do’s and don’ts” are listed. The
policy is also available on the Office of Information
Technology webpage for reference. This is to make sure
that all students are aware of their roles and
responsibilities. One clause clearly states that the video
cannot be taped and distributed for personal purposes or
sold for money. If a student is found guilty they can be
expelled from the University. Annual audits are also
performed to ensure that all of the backup copies are
accounted for.
IV. CONCLUSION AND FUTURE WORK
In this case study we have discussed the different
approaches used by colleges within our University to
protect streaming classes. Though the audience and
distribution methods vary between the Colleges of
Business, Engineering, and the School of Pharmacy the
techniques used to protect the streamed content are largely
the same.
As described earlier in this paper, protection of the
university’s intellectual property is a primary concern.
Unfortunately no delivery system can be considered truly
secure unless the one controls both the transmitting and
receiving computer. Therefore the decision to allow
students to download the course materials becomes a
philosophical one for the university: do the advantages of
allowing students to download materials outweigh the
penalties if a malicious student is able to bypass the
intellectual property controls implemented by the
vendors?
In both cases presented here, the university has decided
that the gains outweigh the costs. Obviously the
protections offered by the software vendors play a role in
this decision. The StarBak system does not natively allow
viewers to download the streaming .asf video files. Hence
the expertise required for a student to upload the course
materials to an alternate site such as YouTube can be
considered high enough to eliminate this as a practical
concern. On the other hand Sony system provides what is
known as progressive streaming. This means that the
computer has to buffer the video stream in a temp file and
users can somehow capture the temp file. Also, other
capture systems allow files to be downloaded to the
computer but the manager can turn this off. In these cases,
the colleges of business and engineering have concluded
that the virtues of distance education outweigh the costs of
having course materials illegally distributed. It is also
worth remembering that the university still enjoys all of
the legal protections afforded by U.S. intellectual property
laws. The videos are copyrighted and any misuse will be
an Intellectual Property right infringement. Students are
fully aware of the repercussions should they upload videos
on YouTube or share it with other universities.
Future work will focus primarily on the upcoming
changes planned for the School of Pharmacy system. The
School of Pharmacy has recently announced plans to
invest more than $25,000 in order to upgrade from version
2.0 to version 3.0 of the Starbak system. Version 3.0
features capabilities and features not found on version 2.0
[3]. How these changes impact the overall security
posture of the University network will be studied with
great detail.
REFERENCES
[1] “Harrison School of Pharmacy” Available:
http://pharmacy.auburn.edu/mobile/index.htm
[2] “Streaming Video: Products: Polycom RSS 2000” Available:
http://www.ivci.com/streaming-polycom-rss-2000.html
[3] “Manage the V3 System” Available:
http://www.starbak.com/solutions/Manage_the_V3_System/manage-
v3.aspx
[4] H. F. Tipton and K. Henry, Official (ISC) 2 guide to the CISSP CBK,
Auerbach Publications.
1
Gwen Greene Applied Information Technology Department
Towson University, Towson, MD 21252, USA
John D‟Arcy Mendoza College of Business, University of Notre Dame,
Notre Dame, IN 46556, USA
Abstract—IS security advocates recommend strategies that shape
user behavior as part of an overall information security
management program. A major challenge for organizations is
encouraging employees to comply with IS security policies. This
paper examines the influence of security-related and employee-
organization relationship factors on users’ IS security compliance
decisions. Specifically, we predict that security culture, job
satisfaction, and perceived organizational support have a positive
effect on users’ IS security
compliance intentions. Our results support the notion that
security culture and job satisfaction lead to increased compliant
security behavior. However, we did not find evidence that
perceived organizational support increases compliant behavior
intentions. Our results do suggest that security culture is a
multidimensional construct consisting of top management
commitment to security, security communication, and computer
monitoring. The findings have implications for information
security research and practice.
Index Terms—Information security, user behavior, security
culture, job satisfaction, perceived organizational support
I. INTRODUCTION
T is recognized among information systems (IS) security
scholars and practitioners that end users are a “weak link” in
security and therefore organizations must develop strategies to
shape user behavior as part of their overall information
security management program. A major challenge for
organizations is encouraging employees to comply with IS
security policies. Many users routinely make a conscious
decision to violate such policies because they want to expedite
their work or increase their own productivity [22]. Others may
violate with more harmful intentions, such as stealing sensitive
corporate data or sabotaging company networks. Regardless of
user intent, research indicates that over half of all information
security breaches stem from users‟ poor (or lack of) IS security
compliance [26][32].
Due to the significance of this issue, a body of literature that
investigates a variety of individual and contextual influences
on compliant user behavior has emerged. One stream is rooted
in deterrence theory, which predicts that the greater perceived
certainty and severity of sanctions for an illicit act, the more
individuals are deterred from that act [11]. The premise of
these studies is that fear of punishment motivates users to
comply with IS security policies. For example, Herath and Rao
[14] found that perceived certainty of detection was positively
associated with users‟ security policy compliance. A separate
stream investigates various non-punishment factors, such as
influence of co-workers, attitudes and beliefs, moral reasoning,
and coping mechanisms, as antecedents to compliant security
behavior. This work incorporates elements from moral
development research models, the theory of reasoned
action/planned behavior, as well as criminological perspectives
including social bond theory, differential association, and
neutralization. Additional research has assessed the influence
work environment factors, such as security culture, as
predictors of compliant behavior [6][10]. D‟Arcy and Greene
[10] found that security culture – operationalized as top
management commitment to security and security
communication – had a positive influence on security
compliance intention among a sample of computer-using
professionals.
Although this prior work is highly informative and
contributes substantial theoretical knowledge to the IS security
domain, we believe that additional research is needed to
provide a more complete understanding of what motivates
users to engage in compliant security behavior. Interestingly,
the vast stream of organizational behavior (OB) literature has
received little attention from IS security scholars. The OB
literature provides a rich source of knowledge on the factors
that contribute to a variety of workplace behaviors, including
behaviors that adhere to corporate policies and regulations. In
the current paper, we draw upon the OB literature and identify
two factors – job satisfaction and perceived organizational
support – that influence security compliance intention. We also
augment D‟Arcy and Greene‟s [10] second-order security
culture construct we an additional factor, computer
monitoring, and posit that security culture, job satisfaction, and
perceived organizational support each have a positive impact
on security compliance intention. The integrated model, shown
in Figure 1, facilitates a comparison of security-related factors
(i.e., security culture) to those of the employee-organization
relationship (i.e., job satisfaction and perceived organizational
support). The results shed light on the factors that motivate
individuals toward compliant IS security behavior in the
workplace.
Assessing the Impact of Security Culture and
the Employee-Organization Relationship on IS
Security Compliance
I
2
Fig. 1. The Research Model
II. SECURITY CULTURE
Organizational culture has been defined as a pattern of
beliefs and expectations shared by organization members [7].
These beliefs and expectations produce norms that shape the
behavior of individuals, groups, or the organization. In a
similar vein, security culture reflects the values and beliefs of
information security shared by all members at all levels of the
organization. Researchers have pointed out that security
culture is an important factor in maintaining an adequate level
of information security in organizations and have even asserted
that only a significant change in security culture can reduce the
number of security breaches experienced [24]. As such,
managers should regard security culture as an important factor
for supporting and guiding their security management
programs [7]. IS security advocates have suggested that
organizations can influence user behavior by cultivating a
security culture that promotes security-conscious decision
making and adherence to security policy [28][29].
Ruighaver et al. [24] argued that security culture is a
multidimensional concept that has often been investigated in a
simplistic manner. In response, D‟Arcy and Greene [10]
opertionalized security culture as a higher (second) order
construct consisting of top management commitment to
security and security communication. In terms of management
support, research has found that beliefs that the decision
makers within organizations have about security, and about the
quality of different processes used to manage security, are
distinguishing characteristics of a security culture [24]. In
short, if information security is clearly important to the
organization, it should be apparent in the actions and
communications of top management.
Education of employees on their roles and responsibilities
related to security is another crucial aspect of security culture
[28]. A well-defined process of security communication
involving education, reminders, and refresher courses increase
employee feelings of responsibility and ownership in decisions
about security, and ultimately lead to a more positive attitude
about security throughout the whole organization [29]. Such
communication efforts manifest themselves in security culture,
which in turn assists in ensuring acceptable actions and
behavioral patterns of end users.
Based on a review of the IS security literature, we identified
a third dimension of security culture: computer monitoring.
Computer monitoring is often used by organizations to gain
compliance with rules and regulations [11]. Within the IS
context, examples include tracking employees‟ Internet use,
recording network activities, and performing security audits.
Monitoring and auditing activities are a means of policy
enforcement and convey the message that the organization is
serious about security and any deviations from mandated
policies and procedures will not be taken lightly [27]. Hence,
computer monitoring is a signal of the organization‟s security
culture.
Although previous studies have not examined security
culture in a holistic manner, extant literature does suggest that
each of our three security culture dimensions impact user
behavior. Hu et al. [15] found that individual behavior toward
information security is influenced by the actions of top
management. IS scholars have also asserted that continuous
education and communication efforts that reiterate the
importance of security help to ensure compliant behavior [28].
Finally, Herath and Rao [14] provide indirect support for the
notion that monitoring increases compliance, as they found a
significant positive relationship between perceived certainty of
detection and security policy compliance intention. Based on
the above, we posit that the stronger the security culture within
the organization, the more likely users will exhibit compliant
behavior. Hence, the following hypothesis:
H1: Security culture is positively associated with security
compliance intention.
III. JOB SATISFACTION AND PERCEIVED
ORGANIZATIONAL SUPPORT
As previously mentioned, we identified job satisfaction and
perceived organizational support as two variables from the OB
literature that have a bearing on decisions to comply with IS
security policy1. Both of these variables have been shown to
influence employees‟ efforts toward achieving organizational
goals, and we therefore reason that this motivational basis will
percolate to the narrower domain of IS security compliance.
Job satisfaction refers to an employee‟s overall sense of
well-being at work [2]. The construct is rather general and
encompasses employee feelings about a variety of intrinsic and
extrinsic job elements. It is believed that employees who
report positive feelings about their employer are likely to
comply with company policies and procedures because they
are highly engaged and see the “big picture” in terms of their
organizational responsibilities. Thus, to the extent that positive
1 The sheer size of the organizational behavior (OB) literature and the
number of variables studied by OB scholars make it impractical to include all
possible constructs that might have a relationship with security compliance
intention. Our examination of the literature led us to pick two of the more
interesting (in our minds) variables to examine. We leave it to future studies
to included additional OB variables.
3
affect is reflected in one‟s job satisfaction, it is likely that more
satisfied employees engage in computing behavior that is in
line with corporate policies and guidelines. Social exchange
theory provides theoretical justification for this assertion. The
theory predicts that people seek to reciprocate those who
benefit them [4]. In an organizational context, if an employee
perceives their work contribution as a positive exchange
relationship in which they feel properly treated by their
organization, they reciprocate through behavior that is positive
and beneficial to the organization [25].
A number of studies provide empirical support for a
relationship between job satisfaction and compliant behavior.
For example, a recent meta-analysis found a positive
correlation between job satisfaction and measures of
performance that included task or in-role performance [16].
Research has also consistently shown that job satisfaction is
related to a variety of organizational citizenship behaviors,
including organizational compliance [19][31]. Based on the
foregoing evidence, we postulate that higher job satisfaction
will manifest itself in an increased tendency toward compliant
security behavior. Hence, the following hypothesis:
H2: Job satisfaction is positively associated with security
compliance intention.
Perceived organizational support (POS) refers to
employees‟ perceptions about the degree to which the
organization cares for their well-being and values their
contributions [12]. POS strengthens employees‟ beliefs that
the organization recognizes and rewards increased
performance or expected behaviors. In return, the employee is
expected to provide dedication and loyalty to the organization
and help the organization reach its objectives [23]. Similar to
the logic used above, social exchange theory provides a
theoretical basis for a relationship between POS and compliant
behavior. When POS is high, a social exchange develops
where the employee may feel compelled to reciprocate the
high level of perceived affective commitment he or she
receives from the organization by engaging in beneficial
behavior. In the IS security context, it is reasonable to assume
that the employee‟s higher level of affective commitment will
lead to a greater adherence to security policies.
The OB literature provides empirical support for this
purported relationship. Settoon et al. [25] cite a number of
studies that found a positive relationship between perceived
organizational support and employees‟ willingness to fulfill
conventional job responsibilities, such as following corporate
rules and guidelines specified in their job descriptions.
Rhoades and Eisenberger [23] found a relationship between
POS and employee job performance in their meta-analysis of
over 70 POS studies. Research has also demonstrated that POS
is associated with increased organizational commitment and
job involvement (i.e., identification with and interest in the
specific work one performs), which in turn are both linked to
increased productivity and prosocial behavior [12][25].
Finally, POS has been shown to be related to lessened
withdrawal behavior and an increased desire to remain with
the organization [23]. Given these empirical findings, the
theoretical predictions of social exchange theory, and the
notion that POS is associated with increased dedication and
loyalty to the organization, we predict a relationship between
POS and IS security compliance intention. Hence, the
following hypothesis:
H3: Perceived organizational support is positively associated
with security compliance intention.
IV. METHODOLOGY
Data were collected using two online surveys that were
administered at separate points in time. The first survey
contained measures for security compliance intention (COMP)
and security communication (COM), as well as demographic
items. The second survey was administered two weeks after
the first and contained measures for top management
commitment to security (TMCS), computer monitoring
(MON), job satisfaction (JS), and perceived organizational
support (POS). The two stage approach allowed us to separate
collection of the independent (with the exception of COM)
from the dependent variable (COMP), thus reducing the
likelihood of common method effects in our results [21].
The survey used several previously validated scales as well
as some original items. TMCS was measured with three items
from Knapp et al. [17]. COM was measured with a six-item
scale developed by the authors, using various items from
Knapp et al. [18]. MON was measured with three items from
D‟Arcy et al.‟s [11] awareness of computer monitoring scale.
JS was measured with five items from Brayfield and Rothe‟s
[5] index of job satisfaction. This scale has been used to
measure overall job satisfaction in numerous studies and has
shown strong psychometric properties. POS was measured
using Eisenberger et al.‟s [12] seven-item scale. The scale
assesses employee valuation of the organization and actions it
might take in situations that affect employee well-being. This
scale has also exhibited strong validity and reliability in prior
studies (e.g., [25]). All survey items were measured on five
point scales ranging from „strongly disagree‟ to „strongly
agree,‟ except for the control variables. The items are
presented in the Appendix.
We reviewed our multi-item measures using criteria for
formative and reflective constructs [20] and determined that
each was reflective. However, we modeled security culture as
a Type II second-order construct – a second order construct
that has formative relationships with its reflective
subconstructs. Recall that security culture is a higher (second)
order construct consisting of three lower (first) order
constructs: TMCS, COM, and MON. At an abstract level, each
of these variables contributes to the overall security culture
construct. However, our literature review (and the inter-
construct correlation results in Table 3) indicates that TMCS,
COM, and MON are distinct dimensions of security culture.
4
A list of 223 potential survey participants was developed
based on the first author‟s professional contact list. This list
consisted of computer-using professionals located in various
organizations throughout the mid-Atlantic region of the U.S.
An email invitation was sent to the list, asking individuals to
complete the first survey. The email also stated that a second
survey would be forthcoming in two weeks. One hundred
forty-six individuals provided usable responses to the first
survey, and 134 provided usable responses to the second.
Participants were asked to provide the last four digits of their
telephone number in each survey so that the researchers could
match their responses. The final sample consisted of 127
usable, matched responses. Table 1 summarizes the respondent
demographic characteristics.
TABLE I
RESPONDENT DEMOGRAPHIC CHARACTERISTICS
V. ANALYSIS AND RESULTS
We used the partial least squares (PLS) structural equation
modeling technique to analyze the data via the SmartPLS 2.0
software package. PLS was chosen because it supports the
testing of formative relationships between higher-order
constructs and their subconstructs. The second-order construct
security culture is estimated using the hierarchical component
approach. In this approach, the higher-order factor is created
using the indicators of its lower-order factors (i.e., TMCS,
COM, and MON). For further explanation, see [8][30].
Following standard procedures, we assessed the
measurement model and then the structural relationships. The
measurement model was assessed through tests of convergent
validity, discriminant validity, and reliability. For convergent
validity, all factor loadings should exceed 0.70 and average
variance extracted (AVE) for each construct should exceed
0.50 [13]. As shown in Table 2, both criteria were met for all
constructs, with the exceptions of TMCS3 (loading of .69) and
POS4 (loading of .63). TABLE 2
LOADINGS, CROSS-LOADINGS, AND AVE‟S
TABLE 3
RELIABILITY AND INTER-CONSTRUCT CORRELATIONS
For discriminant validity, the square root of the AVE for
each construct should be greater than the inter-construct
correlations, and items should load more strongly on their
corresponding construct than on other constructs (i.e., loadings
should be at least .10 than cross-loadings) [13]. As shown in
Tables 2 and 3, these conditions were met for all constructs. It
is worth noting, however, that the JS and POS constructs were
highly correlated at .74. Multicollinearity typically is
considered a serious problem if the correlation between two
variables is greater than 0.8 [3]. We conducted formal tests for
multicollinearity and both constructs had variance inflation
factor (VIF) values less than 3.0, which is below the
conservative cutoff of 3.3 [20]. Further, our discriminant
validity tests revealed that JS and POS were empirically
distinct. Hence, our results corroborate prior research that has
found POS to be related, yet distinct from, JS [23]. We also
note that the correlations between TMCS, COM, and MON are
below the .80 cutoff value commonly used to assess
discriminate validity. This demonstrates the distinctiveness of
content captured by the individual first-order factors,
providing support for the formative nature of security culture.
5
Finally, the reliabilities of all constructs were above the
recommended .70 threshold [13] as shown in the second
column of Table 3.
The test of the structural model includes estimating the path
coefficients, which indicate the strength of the relationships
between the independent and dependent variables, and the R2
value (the variance explained by the independent variables). A
bootstrapping resampling procedure (600 samples) was used to
determine the significance of the paths within the structural
model. We followed a multistage approach and analyzed the
structural model in a manner analogous to hierarchical
regression. This approach facilitates an assessment of the
substantive impact of the two OB variables (JS and POS) on
COMP relative to that of security culture.
The first model consisted of only the control variables and
explained 9% of COMP variance. Next, we add security
culture to the model and the addition of this second-order
factor increased the variance explained to 40%. We then
calculated the effect size by applying the following formula: f2
= (R2 Full Model – R
2 Partial Model)/(1 – R
2 Full Model), which yielded
an effect size of .51. We determined the significance of this
change in R2 through a pseudo F-test in which f
2 was
multiplied by (n – k -1), where n equals sample size and k
equals the number of independent variables. The result of the
pseudo F-test was 62.62 (p<.001). An effect size of 0.02 is
small, 0.15 is moderate, and 0.35 is large [9]. Thus, the
addition of security culture to our control variable-only model
had a large effect size that is significant.
Next, the JS and POS variables were added and they
increased the R2 of COMP from 40 to 45%. This is a small
effect size (.10) but it is significant (12.02, p<.001). Results of
the full PLS model evaluated are shown in Figure 2.
The results show that the three dimensions forming security
culture (TMCS, COM, and MON) each have significant paths.
This indicates that all three first-order constructs make a
unique contribution to the second-order construct, and thus
provide justification for the acceptance of security culture as a
second-order factor. COM had the highest of the three path
coefficients, suggesting a greater importance of this dimension
over TMCS and MON in forming security culture. In terms of
the hypotheses, the results are supportive of H1 as security
culture had a significant positive relationship with COMP (β =
.636, p<0.01). The results also support H2 as JS had a
significant positive relationship with COMP (β = .214,
p<0.05). The relationship between POS and COMP was
significant (β = -.363, p<0.01), but not in the predicted
direction and therefore H3 is not supported. Turning to the
control variables, age (β = .169, p<0.01) was significantly
associated with COMP while gender and industry were not.
Thus, it seems that older employees are more likely to comply
with IS security policies.
Fig. 2. Structural Model Results
As an additional analysis we tested for the moderating
influences of employee type (IT vs. non-IT employee),
industry (IT company vs. non-IT company), and employee
tenure (low tenure vs. high tenure) on the hypothesized
relationships in our model. While we have not committed to a
particular theoretical basis for these moderating effects, it
seems logical that the influences of security culture and our
OB variables on COMP may be contingent on the
aforementioned factors. To explore the moderating influence
of employee type, we split the sample into IT (n=50) and non-
IT (n=77) employee subgroups and ran separate models for
each group. Statistical tests using the approach suggested by
[8] tested the differences in path coefficients between the
subgroups. The results revealed significant differences in the
two groups for the influences of both OB variables on COMP.
For the non-technical subgroup, the relationship between POS
and COMP was nonsignificant while for the technical
subgroup this relationship was negative and significant (β = -
.278, p<0.01). In terms of JS, the relationship between JS and
COMP was positive and moderately significant (β = .123,
p<0.10) for the non-technical group while nonsignificant for
the technical group. These results suggest that job satisfaction
is a somewhat important determinant of security compliance
intention among non-technical employees but has little or no
influence for employees in technical positions. Furthermore,
the unexpected negative relationship between POS and COMP
that was found in the full sample results (Figure 2) seems to be
largely a result of respondents with technical positions.
We followed the same procedure as above and created
subgroups for the respondents from IT companies (n=60) and
non-IT companies (n=67). The only significant differences
between the groups was the relationship from JS to COMP,
with this coefficient being positive and significant (β = .198,
p<0.05) in the non-IT company sample and nonsignificant in
the IT company sample. Hence, it appears that JS is more
influential in determining employee security compliance
intention in non-IT companies compared to IT companies. We
speculate that due to their industry affiliation, employees in IT
companies are more cognizant of security issues and have a
stronger overall sense of the importance of IT security and
6
therefore are more apt toward engaging in compliant security
behavior versus those employees in non-technical industries
that may have less emphasis on IT security and require more
positive affect toward their employer as a means to engage in
compliant security behavior.
Finally, we performed a median split (median = 4 years) on
the tenure variable and created subgroups for the low tenure
(n=65) and high tenure (n=62) employees. As with the industry
subgroups, the only significant differences between the groups
was the relationship from JS to COMP, with this coefficient
being positive and marginally significant (β = .126, p<0.10)
for the low tenure sample and non-significant for the high
tenure sample. This suggests that employees with less time in
the organization are more influenced by their satisfaction
toward their job in terms of engaging in security related
behavior. It may be that higher tenure employees comply with
security policies and procedures out of fear of punishment or
because they have more to lose if they do not comply, whereas
lower tenured employees have less to lose if punished and
therefore make their compliance decisions based on their
feelings of affect toward their job and the organization.
It should also be noted that the influence of security culture
was relatively consistent across the subgroup analyses. Hence,
security culture appears to have a strong influence on security
compliance intention across different employee positions and
industries, and for employees with various years of work
within their respective organizations.
VI. DISCUSSION
This paper introduced and empirically tested a model on the
effect security culture, job satisfaction, and perceived
organizational support on IS security compliance intention. In
doing so, we (1) augmented the second-order security culture
construct with an additional dimension – computer monitoring,
and (2) evaluated the contribution of two employee-
organization relationship variables on IS security compliance
intention. The results suggest that both security-related and
general work environment factors contribute to IS security
compliance. The results also provide some evidence that these
factors are moderated by employee position, industry, and
tenure with the organization.
The observed relationship between security culture and
security compliance intention supports the notion that security
culture is an important factor, and perhaps necessary, for
supporting and guiding information security management
programs. The results also support prior empirical work that
found a relationship between security culture and both in-role
(i.e., formally specified in job descriptions) and extra-role (i.e.,
discretionary) end user security behavior. However, as noted,
extant research has conceptualized security culture as top
management commitment to security and security
communication. We contributed to the content validity of the
security culture construct by identifying computer monitoring
as a third dimension. This is based on the idea that
enforcement activities are a key facet of any information
security culture. Our results point to the unique contribution of
computer monitoring in forming a security culture, while
suggesting that this broader conceptualization of security
culture has a strong influence on users‟ IS security compliance
decisions.
The examination of the OB variables advances our
understanding of the factors that motivate compliant user
behavior in the workplace. We found that higher job
satisfaction is associated with an increased tendency toward
compliant security behavior. This supports the predictions of
social exchange theory in that users will be more apt to engage
in positive and beneficial organizational behaviors if they are
satisfied and perceive their employment relationship as a
positive exchange. The results also extend prior work that
found a link between job satisfaction and desirable employee
behavior to the narrower context of security-related behavior.
In short, satisfied employees appear more likely to comply
with information security mandates.
The results regarding the influence of perceived
organizational support were not as expected and therefore
warrant further discussion. Based on the voluminous OB
literature in this area, we expected perceived organizational
support to have a positive effect on individuals‟ intentions to
engage in compliant security behavior. Instead, we found the
opposite effect; that is, perceived organizational support had a
significant negative association with security compliance
intention. A possible explanation is that employees who
perceive strong organizational support in general may feel that
information security problems will be handled by the IT
department or other functional areas and therefore their own
compliance practices are not critical. This is related to the
concept of risk homeostasis [1]. Risk homeostasis predicts that
people engage in risky behavior when they perceive increased
safety in a particular area (i.e., a false sense of security). Thus,
users may be less vigilant in their security behavior when they
perceive greater organizational support because they equate
this to an abundance of security controls. It may also be that
employees who perceive stronger organizational support feel
that the organization will continue to “have their backs” even
if they fail to comply with security policies and guidelines.
Finally, it is possible that the strong correlation between our
job satisfaction and perceived organizational support construct
measures contributed to the unexpected result. On the other
hand, formal tests indicated that the two constructs were
empirically distinct and their correlation was within the
threshold for establishing discriminant validity. At any rate,
future studies should assess the influences of job satisfaction
and perceived organizational support on security compliance
intention vis-à-vis other employment relationship variables to
add credibility to our findings.
The results of the subgroup analyses suggest that the
relationships between the OB variables and IS security
compliance may be contingent on other factors. Specifically,
the influence of job satisfaction on security compliance
intention is stronger for employees with non-technical
7
positions and for employees in non-IT industries, and for
employees with less tenure in their organizations. Future
research should explore the underlying causes of these
contingent effects and test the purported moderating effects
using stronger empirical tests.
This study has implications for both the research and
practice of information security. From a research perspective,
this study offers an expanded conceptualization of security
culture that accounts for enforcement activities in the form of
computer monitoring. Thus, we offer a more thorough
measurement and analysis of security culture that can be used
in future studies. The current study is also one of the few in the
behavioral information security literature that incorporates OB
variables. As such, we extend the rich body of work on the
employment relationship factors that influence workplace
behavior to the information security realm. From a
methodological perspective, the study improves upon the rigor
of many prior studies through a two-part survey approach that
included a time lag between collection of the independent and
dependent variables (with the exception of COM). This
significantly reduces the likelihood of common method effects
in our data. For practitioners, the results point to the
importance of top management commitment to security,
security communication efforts, and monitoring activities as
drivers of security culture, which in turn contributes to
adherence to security policy. Further, efforts to create a work
environment where employees are satisfied and happy with
their jobs will not only improve the overall quality of life
within the organization, but should also enhance the
organization‟s security posture by increasing security
compliance among the user population.
APPENDIX
REFERENCES
[1] E. Albrechtsen, "A qualitative study of users' view on information
security," Computers & Security, vol. 26, no. 4, pp. 276-289, 2007.
[2] S. Ang, L. van Dyne and T.M. Begley, "The employment relationships
of foreign workers versus local employees: A field study of
organizational justice, job satisfaction, performance, and OCB," Journal
of Organizational Behavior, vol. 24, no. 5, pp. 561-583, 2003.
[3] R.P. Bagozzi, Y. Yi and L.W. Phillips, "Assessing construct validity in
organizational research," Administrative Science Quarterly, vol. 36, no.
3, pp. 421-458, 1991.
[4] P.M. Blau, Exchange and power in social life, John Wiley & Sons, New
York, 1964.
[5] A.H. Brayfield and H.F. Rothe, "An index of job satisfaction," Journal
of Applied Psychology, vol. 35, no. 5, pp. 307-311, 1951.
[6] M. Chan, I. Woon and A. Kankanhalli, "Perceptions of information
security at the workplace: linking information security climate to
compliant behavior," Journal of Information Privacy & Security, vol. 1,
no. 3, pp. 18-41, 2005.
[7] S.E. Chang and C-S. Lin, "Exploring organizational culture for
information security management," Industrial Management & Data
Systems, vol. 107, no. 3, pp. 438-458, 2007.
[8] W. Chin, "Frequently asked questions-partial least squares & pls graph,"
Http://disc-Nt.Cba.Uh.edu/chin/plsfaq.Htm, 2000.
[9] J. Cohen , Statistical power analysis for the behavioral sciences,
second edition, L. Erlbaum Associates, Hillsdale, N.J., 1988.
[10] J. D'Arcy and G. Greene, "The multifaceted nature of security culture
and its influence on end user behavior," in Proceedings of IFIP TC8
International Workshop on Information Systems Security Research,
Cape Town, South Africa, pp. 145-157, May 2009.
[11] J. D'Arcy, A. Hovav and D. Galletta, "User awareness of security
countermeasures and its impact on information systems misuse: A
deterrence approach," Information Systems Research, vol. 20, no. 1, pp.
79-98, 2009.
[12] R. Eisenberger, R. Huntington, S. Hutchison and D. Sowa, "Perceived
organizational support," Journal of Applied Psychology, vol. 71, no. 3,
pp. 500-507, 1986.
[13] D. Gefen and D. Straub, "A practical guide to factorial validity using
pls-graph: Tutorial and annotated example," Communications of AIS,
vol. 16, pp. 91-109, 2005.
[14] T. Herath and H. Rao, "Protection motivation and deterrence: a
framework for security policy compliance in organisations," European
Journal of Information Systems, vol. 18, no. 2, pp. 106-125, 2009.
[15] Q. Hu, T. Dinev, P. Hart and D. Cooke, "Top management
championship and individual behaviour towards information security:
an integrative model," in Proceedings of the Sixteenth European
Conference on Information Systems, Galway, Ireland, June 2008.
[16] S. Kaplan, J.C. Bradley, J.N. Luchman and D. Haynes, "On the role of
positive and negative affectivity in job performance: A meta-analytic
investigation," Journal of Applied Psychology, vol. 94, no. 1, pp. 162-
176, 2009.
[17] K.J. Knapp, "Information security: management's effect on culture and
policy," Information Management & Computer Security, vol. 14, no. 1,
pp. 24-36, 2006.
[18] K.J. Knapp, T.E. Marshall, R.K. Rainer and F.N. Ford, "Managerial
dimensions in information security: A theoretical model of
organizational effectiveness," ISC2 Inc and Auburn University Research
Report, 2005.
[19] J.A. LePine, A. Erez and D.E. Johnson, "The nature and dimensionality
of organizational citizenship behavior: A critical review and meta-
analysis," Journal of Applied Psychology, vol. 87, no. 1, pp. 52-65,
2002.
[20] S. Petter, D. Straub and A. Rai, "Specifying formative constructs in
information systems research," MIS Quarterly, vol. 31, no. 4, pp. 623-
656, 2007.
[21] P.M. Podsakoff, S.B. MacKenzie, J.Y. Lee and N.P. Podsakoff,
"Common method biases in behavioral research: A critical review of the
literature and recommended remedies," Journal of Applied Psychology,
vol. 88, no. 5, pp. 879-903, 2003.
[22] J. Predd, S.L. Pfleeger, J. Hunker and C. Bulford, "Insiders behaving
badly," IEEE Security & Privacy, vol. 6, no. 4, pp. 66-70, 2008.
[23] L. Rhoades and R. Eisenberger, "Perceived organizational support: A
review of the literature," Journal of Applied Psychology, vol. 87, no. 4,
pp. 698-714, 2002.
[24] A.B. Ruighaver, S.B. Maynard and S. Chang, "Organisational security
culture: Extending the end-user perspective," Computers & Security,
vol. 26, no. 1, pp. 56-62, 2007.
8
[25] R.P. Settoon, N. Bennett and R.C. Liden, "Social exchange in
organizations: Perceived organizational support, leader-member
exchange, and employee reciprocity," Journal of Applied Psychology,
vol. 81, n. 3, pp. 219-227, 1996.
[26] J.M. Stanton, K.R. Stam, P. Mastrangelo and J. Jolton, "Analysis of end
user security behaviors," Computers & Security, vol. 24, no. 2, pp. 124-
133, 2005.
[27] D.W. Straub, "Effective IS Security," Information Systems Research,
vol. 1, no. 3, pp. 255-273, 1990.
[28] R. von Solms and B. von Solms, "From policies to culture," Computers
& Security, vol. 23, no. 4, pp. 275-279, 2004.
[29] C. Vroom and R. von Solms, "Towards information security behavioural
compliance," Computers & Security, vol. 23, no. 3, pp. 191-198, 2004.
[30] M. Wetzels, G. Oderkerken-Schroder and C. van Oppen, "Using PLS
path modeling for assessing hierarchical construct models: Guidelines
and empirical illustration," MIS Quarterly, vol. 33, no, 1, pp. 177-195,
2009.
[31] L.J. Williams and S.E. Anderson, "Job satisfaction and organizational
commitment as predictors of organizational citizenship and in-role
behaviors," Journal of Management, vol. 17, no. 3, pp. 601-617, 1991.
[32] T. Wilson, "Understanding the danger within," InformationWeek
Research Report, 2009.
Abstract—Incidents of computer abuse, loss of proprietary
information and security lapses have been on the increase. More
often than not, such security lapses have been attributed to
internal employees in organizations subverting established
organizational controls. What causes employees to engage in such
illicit activities is the focus of this paper. In particular we argue
that increased employee emancipation leads to better protection
of information in organizations. In this paper we present our
argument by reviewing the appropriate theoretical literature and
by developing conceptual clarity of the related constructs.
I. INTRODUCTION
NTERNAL threats from employees have always been
acknowledged as a major source of security breaches in
organizations [37]. A recent report by privacyrights.org on
data breaches indicates a substantial increase of internal
security breaches from 2005 to 2009 that involve various cases
of stealing personal information of internal publics, co-workers
and clients in different organizations such as banking, health,
university and government offices. The breaches cause a lot of
damage to a company‟s computer systems, financial data,
business operations and ultimately, it‟s reputation. These
events are created by the people who are given privileges to
use IS in organizations and subsequently misuse that access
privilege. Stanton et al. [32] describe that regardless of users‟
intent - from naive mistakes to intentional misuse - security
breaches because of lack of compliance can be curtailed. The
importance of the issue, hence, attaches relative importance of
research into insider threats that have consistently focused on
employees‟ compliance behavior to IS policies. IS literature
has a long tradition of studying the management roles in
cultivating a security culture and its impact on intentions to
comply [11][22][31]. The premise of these studies is that
management commitment, communication and enforcement of
power motivates employees to comply with security policies.
For example, a study conducted by Arbodela et al. [4]
involving the identification of antecedents of employees‟
safety compliance behavior discovers that top management
commitment to safety was perceived to be an integral
determinant of safety culture. Additionally, Chan et al. [11]
suggests that upper management practices are found to be
positively related to employees‟ compliance behavior through
mediation of perception of information security climate. Along
the same lines, Spitzmueller and Stanton [31] reveal
that organizational commitment, organizational identification
and attitudes towards technology predict employees‟ intentions
to comply with security policies. An associated stream of research is one that investigates
motivation of management through reward and sanction in
influencing employees‟ compliance behavior. An organization
that is truly dedicated to security will recognize the need for
motivation beyond mere security awareness and will develop
an effective security motivation program along with or as a
part of a continuing awareness effort. It should employ the
most controllable and powerful motivators, rewards and
penalties [27]. A study conducted by Siponen et al. [30] finds
that sanctions have a significant impact on actual compliance
with information security policies. According to Parker [27]
penalties or sanctions often involve loss of favor, perks,
position, or remuneration. One group manager publicly posted
the name of anybody who revealed his or her password and
required everyone in the group to immediately change their
passwords. This produced peer pressure to keep passwords
secret because nobody liked having to learn a new one.
However Pahnila et al. [26], commenting on factors that
explain employees' IS security policy compliance among
Finnish companies, suggest that rewards do not have a
significant effect on actual compliance with IS security policy.
Most of the prior research in internal control towards security
problems deals with protection motivation and cultivating a
security culture. While this stream of research is useful, it falls
short of suggesting why employees subvert security policies in
the first place. Moreover, from an ontological perspective,
previous research orientation is dominated by functionalist and
interpretivistt paradigms [13]. Although these paradigms are
important, information systems security research can also be
studied from alternative viewpoints, viz critical. Based upon
the notion that critical humanist study of information systems
security is underdeveloped, we attempt to explain the issue of
employees‟ intention to comply with security policy by
arguing that lack of emancipation of an individual is a
significant cause for security lapses.
Yurita Abdul Talib and Gurpreet Dhillon School of Business, Virginia Commonwealth University
301 W Main Street, Richmond, VA
{abdultalibyy, gdhillon} @vcu.edu
d
Invited Paper: Employee Emancipation and
Protection of Information
I
II. WHO IS AN EMANICIPATED EMPLOYEE?
The concept of emancipation was established by Critical
Social Theorists (CST) as an alternative to traditional research
approaches. The fundamental goal of CST is improvement of
human conditions, which takes into account the human
construction of social forms of life and the possibility of their
recreation [25]. The critical theorists seek to emancipate
people; they are concerned with finding alternatives to existing
social conditions. More specifically, with emancipation, the
organization actors have the capability to transform
organizational conditions.
Habermas‟s Theory of Communicative Action is an
extension of CST with a broader notion of rationality.
Habermas developed the concept of "communicative action",
defined as "the type of interaction in which all participants
harmonize their individual plans of action with one another
and thus pursue their illocutionary aims without reservation"
[16]. According to this perspective, in order to overcome
social crises, it is necessary to counterbalance purposive
rationality by bringing communicative rationality back into
play. Hence the framework highlights the importance of
developing a society based on free, undistorted communication
in order to prevent colonization and technization of the life
and world by power and money [16].
The study of human emancipation has been long established
in the field of social studies. And among the well-known
studies are those on 'women emancipation'
[1][2][8][12][18]21]. The term 'women emancipation' is used
to denote equalization of opportunity structure in which
women are to gain equal status as men as a result of balancing
of gender roles [21]. Women now are no longer stuck in the
kitchen with baby diapers, but are involved actively in social
structures. The prominent benefit of equality of gender is in
education, in which for a long time in history, women were not
given opportunity to obtain an education. Education opens up
many opportunities for women, while eliminating ignorance
and silent suffering. Women are now emancipated enough to
join the labor force in which they can be found amongst the
most powerful politicians, scholars, CEOs of companies and
holding many more jobs that were earlier monopolized by
men. This movement of equality in gender roles or
emancipation allows women to discover their suppressed
dignity and potential. Though some proponents of women
emancipation see it as a stimulus for crime committed by
women and delinquent behaviors, most support the idea as a
means of manifestation of democratic ideals.
Hirschheim and Klein [17] purport that “Information
systems as social communication systems have potential of
freeing employees from repressive social and ideological
conditions and thereby contributing to the realization of human
need” (p.87), i.e., information systems facilitate emancipation.
This notion is supported in information systems that are a basis
of poor information access, particularly via traditional method
of information dissemination and sharing in a large and
distributable organization [9]. However, due to the fact that
'information is power' [35], organizations impose security
mechanisms in terms of limiting employees‟ access to
information systems in order to preserve management power.
The inequality of power in information access creates
oppressed individuals, which significantly contradicts the
earlier view that information systems should free employees
from oppressive conditions [17].
Given the aforementioned, in this paper we aim to stimulate
the idea of employees‟ emancipation in information access in
organizational information systems. Ultimately the central
objective is to relate employees‟ emancipation and their
compliant behavior. In cultivating the idea, we base our
conception upon the concept of women emancipation, in which
managers and employees are given equal right to access an
organization‟s information. The transformation in information
sharing and decentralization of power would then dismiss the
power structure that alienates the employees and makes it
difficult for them to undertake their work and get involved in
decision-making processes. Emancipation therefore provides
equality between managers and workers, hence allowing for
sharing of responsibilities and freedom to take decisions.
Responsibility for successful decisions‟ results in employees
becoming more motivated and hence increases their morale
and develops a feeling of ownership. In turn, the ownership
feeling makes employees feel responsible towards protecting
the assets of an organization.
III. HOW EMANCIPATION FACILITATES
INFORMATION PROTECTION?
In an information systems environment, there are only a few
individuals (usually the management), who have a privileged
access to information and hence organizational power. This is
usually at the expense of other people in the organization (i.e.
employees). Since information is power [35], and only a
privileged few have access, it leaves a vast majority as being
powerless. Hence the notion that power 'creates' oppressed
individuals in which it stops them from carrying out what they
would freely choose to do otherwise, is realized [7]. Limiting
employees‟ access to information not only creates inequality of
power but also restricts the involvement of the oppressed in
the decision making process. Ultimately, these employees get
alienated, disgruntled and generally do not feel that they are an
active participant in organizational affairs.
Alienation of employees as the result of the power structures
can be represented by four dimensions, namely powerlessness,
meaninglessness, isolation and self estrangement [33].
Correspondingly, the feelings of powerlessness as a result of
being controlled by others and isolation, that leads to a lack of
sense of belonging, diminishes the worker‟s commitment
towards the organization [33]. As a result, employees who feel
alienated from their employer are less motivated to commit
and more likely to take decisions that are unethical. This in
turn leads to non-compliance with security policies. A
combination of these circumstances makes an organization
susceptible to security breaches.
In this paper, the focus of employees‟ emancipation is
towards information access in an organization's information
systems. We define emancipation as freeing the employees
from the power structure by increasing the scope and depth of
their information access. This, we argue, leads to
decentralization of power between employees and managers,
which subsequently provides employees with the authority to
be involved in the decision making processes. Information
sharing is an instrument in eliciting employees‟ involvement
and building trust [19]. Based upon decision making
involvement and the culture of trust that an organization
creates, it can foster greater emotional buy-in from
employees.
As an example, imagine that employees are allowed to share
information about financial status during a crisis, at which time
the management often hoards and restricts access to such
information. However the employees might be able to help
with suggestions. Being at the operational level, employees
know better as to what specific actions affect the overall
business. They are also more likely to offer valuable ideas on
how the operations can be improved. Since meaningful
suggestions from employees are more likely to be adopted, it
builds a situation where the employees feel valued by the
organization. That in turns can help build morale, giving
employees a greater sense of worth and emotional ownership
in the company.
This sense of ownership or buy-in triggers a sense of
responsibility for the entity [14]. When employees feel
responsible for the entity, they will protect what belongs to the
entity - in our case the information systems and information
that they handle. This is logically supported by the assertion
that 'rational' people will not destroy or take unnecessary
action against something that belongs to him. As much as
possible these people will protect what they are responsible
for. This assertion is fundamental to our argument that
emancipation of employees, by allowing them access to
information and participation in the decision making process,
leads to a sense of ownership and responsibility. This will
form the basis for ensuring complete support and adherence of
all employees to organizational security policies.
IV. ON THE NATURE AND SCOPE OF “RESPONSIBLE
BEHAVIOR” FOR INFORMATION PROTECTION
Employees‟ sense of ownership towards the information,
information systems and hence the organization itself may
have a positive impact on employees‟ behavior. It has been
argued that ownership creates a sense of responsibility for an
object [14] which initiates behaviors of protecting the owned
object [35].
Bear, Manning and Izard [15] purport that responsible
behavior in obedience and compliance to a rule entails self-
motivation and self-guidance, and is not merely a response to
external supervision, rewards and punishment. Hence, prior
research on employees‟ compliance behavior to security policy
that focuses on these external factors is not the sole contributor
for building responsible behavior in employees‟. While
embedding these factors, which are merely enforcement of
power, is empirically effective, over time their significance
will be eroded. These external factors change over time and
are impossible for employees to follow because of the
constantly shifting direction to conform to the influences [5].
Therefore we believe that the good behavior of the employees
on policy compliance is only temporary, if it is present at all.
On the other hand, responsibility behavior of employees,
which derives from employees' cognitive and emotional factors
[15], would deem to be more effective in shaping their
behavior towards compliance. With responsible behavior,
people will act accordingly because they perceive that they are
the cause of their own behavior, in which case, if they do not
comply with the policies, they are responsible for that as well,
not the other people or some other external factors. People
with a responsible behavior will act the way they should
whether anyone is watching or not and also remain aware of
the consequences of their behavior for the welfare of others.
In summary, as stated previously, the emancipation of
employees allows them to acquire a feeling of ownership
towards the organization (object). This is achieved by
providing an equality of power in term of information access
and consequent involvement in the decision making process.
This ownership signifies that both employees and managers
are entitled to use and engage with the object, and hence be
held responsible by each other in protecting the owned object.
Based upon this assumption, we postulate that employees will
protect the organization from any harmful acts by performing a
„right‟ responsible behavior in accordance with organizational
security policies.
V. A CONCEPTUAL MODEL OF EMANCIPATION AND
INFORMATION PROTECTION
Emancipation refers to freeing employees from oppressive
conditions, hence enabling them to realize their full potential
[3]. According to Alvesson and Willmott [3], the concept of
emancipation in organizations is described as “freeing
employees from unnecessarily alienating forms of work in
organization” (p.433). Ultimately the central focus of
emancipation lies is providing communication discourse in
which all level of employees are able to access the same
amount of information and are involved with decision making
processes, share responsibility and hence promote democracy
in organizations. In the same vein Hirschheim and Klein [17]
purport that emancipation can be practiced in organization
through involvement in decision making process, which in turn
reduces the power of management and increases employees‟
responsibility. Similarly, Sashkin [28] identifies the need for
employers to fulfill employees‟ basic human needs at the
workplace for autonomy, meaningfulness of work and
decreased isolation. Failure to provide those needs, according
to Sashkin [28], is unethical as it creates physiological and
physical harm to the employees. In relation to the previous
literatures, we propose that while employees‟ involvement in
decision making process is vital, it is not achievable if they are
not emancipated in accessing the information. Our argument is
based on current practices of an organization in the digital age
wherein most information, including that for decision making
purposes, are inputs, processed and delivered via vastly
controlled computer based information systems. Without
substantial and suitably structured access privileges for the
information, it is rather difficult for the employees to be
involved in and contribute to the process. A summary of our
postulated relationships of emancipation and information
protection are presented in Figure 1.
Emancipation of individuals has become an increasingly
important method of increasing employees‟ creativity and
productivity and researches have directed increased attention
towards emancipation effectiveness. Research results
consistently contribute to the notion that giving the employees
involved in the tasks the authority to take decisions, makes
them more creative and motivated to aim for successful
production decisions. In a security context, the aim is towards
protection of the information, particularly in compliance with
security policies. Although previous studies in IS have
examined the positive consequences of emancipating
employees in information system development, none has made
an attempt to study emancipation of employees for access to
information. Based on the above, we posit that by
emancipating employees with respect to information access,
employees are more likely to comply with an organization's
security policies directly or indirectly.
Fig. 1. Conceptualizing employee emancipation and protection of
information
As previously mentioned we postulate that emancipation can
also have an indirect relationship with protection behavior.
One of the indirect routes is its influence on employees‟
commitment to the organization. Employees‟ commitment is
defined as the feeling of desire, need or obligation to remain in
an organization [23]. As employees are given more power
towards information access and deliberate decision making
authority, it increases their commitment towards the
organization. There is support in the literature for this
contention, which claims that participation in decision making
increases employees‟ organizational commitment [10][30]
whereas alienation of employees from an organization will
diminish their commitment [33].
In information systems, the concept of emancipation has
largely been investigated within the realm of information
systems development (ISD), particularly through the
application of both Habermas‟s Theory of Communicative
Action and Critical Social Theory. Cecev-Kecmanovic and
Marius [9] address effectiveness of emancipation in their 15
years longitudinal case study of development of Information
System for Information Dissemination (ISID) in Colruyt
Company in Belgium. They discover that with the amount of
increased power given to the employees, they are less alienated
and oppressed, leading to increased individual commitment.
Similarly Kanungo [20] finds that impact of emancipatory
roles of ICT development in rural areas suggests that by
involving villagers with the development process, they become
more committed towards using the systems. Regardless of the
context of the organization, employees‟ commitment will
direct an individual‟s effort toward achieving organizational
goals [24].
Another stream of research, as introduced previously,
studies the achievement of an organization‟s security goals
through a sense of ownership and responsible behavior.
Previous studies provide substantial evidence that employees‟
involvement in decision making process creates a stronger
sense of ownership or identity [6][29]. With the empowerment
of being involved in decision making process, employees will
be able to contribute more valuable ideas to improve
operations since they are attached directly at the operational
level. Subsequently, their meaningful suggestions are more
likely to be accepted and adopted. This in turns strengthens
motivation by providing employees with the opportunity to
attain intrinsic rewards from their work, such as a greater
feeling of being valued. This helps in inculcating and
increasing a sense of ownership towards the organization.
Our definition of sense of ownership is borrowed from van
Dyne and Pierce [34] which defines it as “psychologically
experienced phenomenon in which an employee develops
possessive feelings for the target” (p. 439). The feeling of
ownership makes the employees feel they own the
organization, in other words- they have a feeling that the
organization belongs to them, which is fundamentally different
from the feeling of their need to remain in an organization or
organizational commitment [34]. If employees feel that they
are part of the business processes, they become much more
attached to their work. As a result employees become more
conscientious of their work and how it affects the organization.
Thus, the feeling of ownership initiates a sense of
responsibility for the entity [34].
Being responsible, according to Bear et al. [15], means the
ability to take decisions that are morally and socially right.
This conveys the notion that employees will act accordingly
because they realize that they are accountable for their own
behavior, regardless of whether they are being monitored or
not. Bear et al. [15] contend that responsible behavior that is
derived from social cognition and individual emotion is more
important than responsible behavior constituted by external
TABLE I
PROPOSITIONS POSTULATING RELATIONSHIPS BETWEEN
EMANCIPATION AND INFORMATION PROTECTION
Proposition Description
1 Emancipation will positively influence information
protection, both directly and indirectly through affective
commitment, sense of ownership and responsible
behavior.
2 Emancipation is positively associated with employee
commitment to an organization
3 Employee commitment to an organization is positively
associated with employee protection of information
4 Emancipation is positively associated with an employee's
sense of ownership / belonging leading to information
protection
5 Employee's sense of ownership / belonging is positively
associated with responsible behavior leading to
information protection
6 Employee responsible behavior is positively associated
with protection of information
factors. In their study, they gather evidence on how
responsible behavior of a student not only benefits the
individual but also the members of a school community. The
behavior promotes positive effects such as academic
achievement and self-worth. Motivated from this notion, we
argue that employee's feelings and thoughts (sense of
ownership / feeling of being valued /feeling of importance) are
the primary determinants that explain how individuals become
responsible employees. Embedded responsible behavior
influences them to promote positive attitude such as protection
of information. A summary of our propositions, as discussed in
this section are presented in Table I.
VI. CONCLUSION
In this paper we have introduced the concept of employees‟
emancipation and its relationship with organizational
commitment, sense of ownership, responsible behavior and
protection of information within an organization. While many
researchers have commented on an employees‟ behavioral
compliance with security policies and laws and regulations,
our research introduces fresh insight by establishing a
relationship between emancipation and information protection.
Rather than be enshrouded in positivist or interpretivist
conceptions, in this research we introduce a critical theorist
orientation for the study of information protection. We have
argued that protection of information in an organization is
influenced by the human cognitive and emotional feelings of
individuals, which are derived from the emancipation of an
employee from organizational power structures that govern
information access. Hence, this paper lays a foundation for
further theoretical and empirical research on the protection of
information.
REFERENCES
[1] F. Adler, Sisters in Crime: The Rise of the New Female Criminal. New
York: McGraw-Hill. 1975
[2] F. Adler, “The interaction between women‟s emancipation and female
criminality: a cross-cultural perspective”, International Journal of
Criminology and Penology, 5, 1977, 101-112, 1977.
[3] M. Alvesson and H. Willmott, “On the Idea of Emancipation In
Management And Organization Studies”, Academy of Management
Review, 17(3), 432-464, 1992.
[4] A. Arboleda, P.C. Morrow, M. R. Crum, and S.C. Mack, “Management
Practices as Antecedents of Safety Culture within The Trucking
Industry: Similarities And Differences By Hierarchical Level”, Journal
of Safety Research 34, no. 2: 189-197, 2003.
[5] A. Bandura, “Social cognitive theory: an agentic perspective”, Annual
Review Of Psychology, 521-26, 2001.
[6] B. Benkhoff, “Ignoring Commitment Is Costly: New Approaches
Establish the Missing Link Between Commitment and Performance”,
Human Relations, 50(6), 701-726, 1997.
[7] K. Booth, “Security and emancipation”, Review of International Studies,
17(4), 313-26, 1991.
[8] S. Box & C. Hale, “ Liberation/emancipation, economic marginalization,
or less chivalry: The relevance of three theoretical arguments to female
crime patterns in England and Wales, 1951–1980”, Criminology: An
Interdisciplinary Journal, 22(4), 473-497, 1984.
[9] D. Cecez-Kecmanovic, M. Janson & A. Brown, “The rationality
framework for a critical study of information systems”, Journal of
Information Technology (Routledge, Ltd.), 17(4), 215-227, 2002.
[10] A.L. Pearson and C. Duffy, “Contexts Commitment and Job
Satisfaction: A Study in Australian and Malaysian Nursing The
Importance of Job Content and Social Information on Organizational”,
Asia Pacific Journal of Human Resources 1999; 36; 17, 1999.
[11] M. Chan, I. Woon and A. Kankanhalli “Perceptions of Information
Security in the Workplace: Linking Information Security Climate to
Compliant Behavior”, Journal of Information Privacy & Security 1, no.
3: 18-41, 2005.
[12] R. Deming, Women: the new criminals. Nashville: Thomas Nelson,
1997.
[13] G. Dhillon and J. Backhouse, “Current directions in IS security
research: towards socio-organizational perspectives”, .Information
Systems Journal, 11(2), 127, 2001.
[14] L. Furby, “Possession in humans: an exploratory study of its meaning
and motivation”, Social Behavior and Personality, 6, 49–65, 1978.
[15] G.G. Bear, M. A. Manning & C. E. Izard, “Responsible Behavior: The
Importance of Social Cognition and Emotion”, School Psychology
Quarterly, 18(2), 140-157, 2003.
[16] J. Habermas, The theory of communicative action, Volume 2. Boston,
MA. Beacon Press, 1987.
[17] R.A. Hirchheim, and H.K. Klein, “Realizing emancipatory principles in
information systems development: The Case for ETHICS”, MIS
Quarterly, 18(1), 83-109, 1994.
[18] J. James and W. Thornton, “Women's Liberation and The Female
Delinquent”, Journal of Research in Crime & Delinquency, 17(2), 230,
1980.
[19] G. M. Kandathil and R. Varman, “Contradictions of Employee
Involvement, Information Sharing and Expectations: A Case Study of an
Indian Worker Cooperative”, Economic and Industrial Democracy
28(1): 140-174, 2007.
[20] S. Kanungo, “On the emancipatory role of rural information systems”,
Information Technology & People, 17(4), 407-422, 2004.
[21] S. Karstedt, “Emancipation, Crime and Problem Behavior of Women: A
Perspective from Germany”, Gender Issues,18(3), 21, 2000.
[22] K.J. Knapp, T.E. Marshall, R.K. Rainer and F.N. Ford, “Information
security: management's effect on culture and policy”, Information
Management & Computer Security 14, no. 1: 24-36, 2006.
[23] J. Meyer and N. Allen, Commitment in the workplace: Theory,
research, and application. Thousand Oaks, CA US: Sage Publications,
Inc. 1997.
[24] J. Meyer and C. Smith, “HRM Practices and Organizational
Commitment: Test of a Mediation Model”, Canadian Journal of
Administrative Sciences (Canadian Journal of Administrative
Sciences), 17(4), 319-331, 2000.
[25] O.K. Ngwenyama, “The Critical Social Theory Approach to Information
Systems: Problems and Challenges,” in Information Systems Research:
Contemporary Approaches and Emergent Traditions, H,E, Nissen, H,
Klein, and R, Hirschheim (eds,), North-Holland, Amsterdam, pp. 267-
278, 1991.
[26] S. Pahnila, M. Siponen and A. Mahmood, “Employees‟ Behavior
towards IS Security Policy Compliance”, Proceedings of the 40th
Hawaii International Conference on System Sciences 2007.
[27] D.B. Parker, “Motivating the Workforce to Support Security”, Risk
Management (00355593) 50, no. 7: 16-19, 2003.
[28] M. Sashkin, “Participative management remains an ethical imperative”,
Organizational Dynamics 14, no. 4: 62-75, 1986.
[29] B. Scott-Ladd, A. Travaglione and V. Marshall, “Causal inferences
between participation in decision making, task attributes, work effort,
rewards, job satisfaction and commitment”, Leadership & Organization
Development Journal, 27(5), 399-414, 2006.
[30] M. Siponen, S. Pahnila and A. Mahmood, “New Approaches for
Security, Privacy and Trust in Complex Environments”, IFIP
International Federation for Information Processing, Volume 232,
133–144, 2007.
[31] C. Spitzmüller and J.M. Stanton, “Examining employee compliance
with organizational surveillance and monitoring”, Journal of
Occupational & Organizational Psychology 79, no. 2: 245-272, 2006.
[32] J.M. Stanton, K.R. Stam, P. Mastrangelo, and J. Joiton, “Analysis of
end user security behaviors”, Computers & Security 24, no. 2: 124-133,
2005.
[33] G. Tonks and L. Nelson, “HRM: A Contributor To Employee
Alienation?” Research & Practice in Human Resource
Management, 16(1), 1-17, 2008.
[34] L. van Dyne and J. Pierce, “Psychological ownership and feelings of
possession: three field studies predicting employee attitudes and
organizational citizenship behavior”, Journal of Organizational
Behavior, 25(4), 439-459, 2004.
[35] S. Zuboff, In the Age of the Smart Machine. New York, Basic Books,
1984.
[36] A. Melek and M. MacKinnon. (2006). Deloitte global security survey.
[Online]. Available:
http://www.deloitte.com/dtt/cda/doc/content/us_fsi_150606globalsecuri
tysurvey(1).pdf
Abstract—We present an instructional suite of labs that can
help in teaching security courses. Our labs are publicly available
for use. We have used these labs in teaching security courses and
also helped several other instructors adopt these labs to suit their
style of teaching. Based on this experience, we present some
guidelines for using these labs in security courses. Furthermore,
we also describe the results of using these labs among students.
Index Terms—Security Education, Laboratory Exercises,
Hands-On Learning
I. INTRODUCTION
T is widely known that learning-by-doing can significantly
enhance student’s learning in many disciplines [1], [5], in-
cluding computer security education [3]. Eight years ago,
when we started looking for well-design hands-on projects for
our security courses, we could not find many. Although we did
find a small number of good individual laboratories [2], [4],
[6], there were two problems in using them. First, the coverage
of security principles was quiet narrow. Second, the lab
environments were varied and as we intend to use these labs in
a single course, students would have to spend a significant
amount of time to learn to use the underlying environment.
Motived by the need for better, coherently-designed, and
well-supported lab exercises for security education, we started
our journeys in 2002. Our objective was to develop a suite of
labs that cover a wide spectrum of principles, ideas, and
technologies that are essential for teaching computer security.
We call these labs the SEED labs (SEED stands for Security
EDucation). Our design particularly focused on two aspects:
the lab environment and the lab contents.
A. Low-cost and Consistent Environment
First, we wanted to have a unified lab environment that is
consistent for different lab exercises, so students do not need
to learn a new lab environment for each lab, because learning a
new environment is not easy. Second, the environment used
for these labs must be affordable to enable wider adoption.
With these constraints in mind, created a lab environment
comprising the open-source Linux Operating system and a
This work was partially sponsored by Award No. 0618680 from the United
States National Science Foundation (NSF).
number of open-source software tools. The lab environment
can be created on student’s personal computers or
department’s general computers with the use of a virtual
machine (VM) software. There are free VM software, such as
VirtualBox, and even the commercial product VMware is free
for educational use through VMware’s academic program.
B. Lab Contents with Wide Coverage
Unlike some traditional courses, such as operating systems
and compilers, there is no common consensus on how security
should be taught, and what contents should be covered.
Therefore, we decided not to base our labs on a particular
course. Instead, we decided to develop labs that can cover a
wide spectrum of security principles. Although the style and
focus of security courses could vary with instructors, the
fundamental security principles that all instructors would like
to teach do not significantly vary. Because the labs cover a
range of security principles, instructors could choose the labs
from our list to fit their course based on the security principles
they want to cover. Adoption of our lab exercises do not
require any changes in the course structure.
After 8 years’ working on the project (funded by two NSF
grants), we have developed over 25 labs, most of which
have been used and tested by courses at our own university;
several of them were also used by instructors from other
universities. The labs are available from our project web site:
http://www.cis.syr.edu/_wedu/seed/. Since 2007, the labs have
been downloaded for over 40, 000 times from the computers
outside of our university network. The labs are broadly
categorized into three classes:
1) Vulnerability and attack exercises
The objective of these exercises is to illustrate and help the
student learn a vulnerability in detail. Each of these exercises
typically has four parts. First, students will understand the
vulnerability based on simple examples and identify them in
either a real program or a synthetic example. Second, students
will construct real attacks to take advantage of the
vulnerability. Third, students will fix the vulnerabilities and
also comment on prevailing mitigation methods and their
effectiveness.
Enhancing Security Education
with Hands-On Laboratory Exercises
Wenliang Du1, Karthick Jayaraman
1, and Noreen B. Gaubatz
2
Department of EECS1, Office of International Research and Assessment
2
Syracuse University, Syracuse, NY 13244
{wedu, kjayaram, nbgaubat}@syr.edu
I
2) Design and implementation exercises
The objective of these exercises is to help the student
understand the concepts of secure system development.
Students are expected to design a system that provides a
specific security functionality. The exercise outlines clear
objectives and students would explore various choices for
fulfilling the objectives, analyze the impact of the choices on
security and performance, make appropriate design decisions,
and finally implement and demonstrate their system.
3) Exploration exercises
The objective of these exercises is to coerce the students to
explore or play with an existing security functionality.
Exploration labs are like a “guided tour” of a system, in which,
students can “touch” and “interact with” the key components
of a security system to either learn or question the underlying
security principles that the system is based on.
C. Experiments and Evaluation
Over the last 7 years, we used the SEED labs in two
graduate and one undergraduate security courses. We have
been revising those labs based on student feedback. Now, most
of the labs are quite stabilized, and are ready for use by other
instructors. So far, fifteen other universities have informed us
that they have used some of our SEED labs in their courses.
To evaluate the effectiveness of our labs, we collected survey
data from students for each of the labs they used. The results
are quite encouraging. We will present our evaluation results
in this paper.
II. DESCRIPTION OF LABS
In this section, we provide a brief description of our SEED
labs. Since the actual lab description for each lab ranges
between 2 pages to 10 pages, we cannot include them in this
paper. We will only describe a selected few. Readers can find
the detailed lab descriptions from our web page
http://www.cis.syr.edu/_wedu/seed/. Table I contains the list of
labs.
A. Vulnerability and Attacks Labs
All the vulnerability and attack labs are structured to let the
student learn the vulnerability in detail by creating an exploit
for a program with known vulnerabilities. Furthermore, the
students are asked to comment on the prevailing mitigating
methods and their effectiveness. Table I contains all 12
vulnerability and attack labs. Broadly, we have three
categories of vulnerabilities, general software vulnerabilities,
network protocol vulnerabilities, and web application
vulnerabilities. We describe the nature of the labs in each of
the categories. The detailed lab description can be obtained
from our web site. From our experience, the vulnerability labs
are heavily used by several instructors.
General Software Vulnerabilities
We have lab exercises focusing on general software
vulnerabilities such as buffer overflows, format-string
vulnerabilities, race conditions, etc. Latest versions of Linux-
based systems have several features for countering these
vulnerabilities. Therefore, where necessary, the lab exercises
provide information on how to disable the vulnerabilities in
order to perform the lab exercises.
Network Protocol Vulnerabilities
In this category, we have two labs, namely TCP/IP attacks and
DNS pharming. In the TCP/IP attack lab, students discover
and exploit vulnerabilities of the TCP/IP implementations. In
the DNS pharming lab, students understand how DNS
pharming attacks could be used to manipulate the name
resolution process of trusted web sites to compromise the user
privacy.
Web Application Vulnerabilities
We have labs focusing on web-based vulnerabilities such as
cross-site scripting, cross-site-request forgery, clickjacking,
etc. The lab exercises for these labs involve exploiting a web
application with known vulnerabilities. We have setup some
web applications inside our pre-configured virtual machine
that students will use for these labs. Therefore, instructors need
not maintain any web servers or install or configure
applications for using these labs.
B. Exploration Labs
In the exploration labs, students both learn and critique the
design and implementation of an existing security component.
We have four labs in this category:
Packet Sniffing and Spoofing Lab
In this lab, students learn how packet sniffing and spoofing
tools are implemented. Students will start with some simple
tools, understand them by reading their source code and using
them, and modify them to implement some interesting variants.
Linux Capability Lab
The objective of this lab is to train students to use the
capability system implemented in Linux and also gain
insights on how the capability system is implemented in the
operating system.
Web Browser Access Control Lab
The objective of this lab is to help the student understand
and critique the same origin policy commonly implemented in
all web browsers for access control.
Pluggable Authentication Module
The objective of this lab is to illustrate the use of the
flexible authentication technique called pluggable
authentication module (PAM) implemented in Linux. PAM
provides a way for developing programs that are independent
of the authentication scheme.
C. Design Labs
Each of the design labs provides clear functional
requirements for implementing a system. Furthermore,
students are also provided supporting material and documents
for helping them. To help the students get started, some design
choices are outlined. The students are first asked to provide a
detailed design of the system describing why they choose
particular design choices and how they impact the security and
performance of the system. The faculty or TA could provide
feedback on their design, before the student implement the
system. Currently, we have nine design labs. We provide a
brief description of these labs as follows:
Linux Firewall Lab and Minix Firewall Lab
In this lab exercise, students are expected to implement a
simple packet filtering firewall. The firewall should provide an
interface to the user to enable configuration. Each firewall rule
describes the types of packets, in terms of their properties such
as source IP address, protocol, etc., that will be allowed or
blocked. The design labs are available in both Linux and
Minix operating systems. For Minix, students have to
modify the Minix kernel to implement the firewall. For
Linux operating system, students will use the kernel
loadable module mechanism and the netfilter hooks to
implement the firewall.
IPSec Lab
In this lab, students will implement a simplified version of
the IPSec protocol for the Minix operating system. The lab
tasks have been simplified to facilitate the students to complete
the lab exercise within four to six weeks. Students need to use
their IPSec implementation to construct a Virtual Private
Network (VPN).
VPN Lab
VPN is a widely used security technology. VPN can be built
upon IPSec or Secure Socket Layer (SSL). These are two
fundamentally different approaches for building VPNs. In this
lab, we focus on the SSL-based VPNs. This type of VPN is
often referred to as SSL VPNs. The learning objective of this
lab is for students to master the network and security
technologies underlying SSL VPNs. The design and
implementation of SSL VPNs exemplify a number of security
TABLE I
THE COVERAGE OF SEED LABS
Legend: ST - Systems, NW - Networking, PG - Programming, SWE - Software engineering. G - For Graduate students only,
UG - For both Undergraduates and Graduates. AU - Authentication, AC - Access control, CG - Cryptography SC - Secure coding,
SD - Secure design.
principles and technologies, including cryptography, integrity,
authentication, key management, key exchange, and Public-
Key Infrastructure (PKI). To achieve this goal, students will
implement a simple SSL VPN for Ubuntu.
Role-Based Access Control (RBAC) Lab
In this lab, student will implement an integrated access-
control system that uses both capability and RBAC access
control mechanisms. The lab tasks require students to modify
the Minix kernel to implement the proposed system. •
Capability lab: In this lab, students will design and implement
a capability-based access control system in Minix. Similar to
the RBAC lab, the lab tasks will involve modifying the Minix
kernel. The capability-based access control system would co-
exist with the file-based permission used in Minix.
Encrypted File System Lab
In this lab, students will design and implement an encrypted
file system for Minix. A key requirement of the design is to
ensure that EFS is transparent to the applications. From the
users perspective, there should be no more work other than
mounting the new filesystem to be able to use the new file
system.
Set-random UID Lab
In this lab, students will implement a simple sandbox for
executing untrusted programs. In the simple sandbox,
programs are assigned randomly generated uid that do not
exist in the system. Because the random uid do not own files,
the access permissions of the executed program are limited.
Address Space Randomization Lab:
In this lab, students will implement methods for
randomizing the heap and stack of the Minix operating
system.
III. COVERAGE OF SEED LABS
A. Principle Coverage
Regardless of how instructors teach computer security and
in what contexts (e.g. networking, operating system, etc.), one
thing is for sure: they should cover the principles of computer
security. In civil engineering, when building bridges, there are
well-established principles that need to be followed. Security
engineering is no different: in order to build a software system
that is intended to be secure, we also need to follow principles.
Regardless of how computer security is taught, the
fundamental principles that most instructors cover are quite the
same, even though the principles might be covered in different
contexts.
The definition of “security principles” is interpreted
differently by different people: some interpret it as software
engineering principles, such as the principle of least privileges;
some interpret it as access control, authentication, etc. To
avoid confusion, we use the following definition:
A computer security principle is an accepted or professed
rule of action or conduct in building a software or hardware
system that is intended to be secure.
We broadly categorized the fundamental computer security
principles into the following classes: Authentication (AU),
Access Control (AC), Cryptography (CG), Secure Coding
(SC), and Secure Design (SD). Table I also shows the security
principles covered by each of the SEED labs.
B. Courses Coverage
After studying a number of security courses taught at
different universities and colleges, we identified several
representative types of courses, and provide suggestions
regarding what SEED labs are appropriate for these courses
(Table I).
System-focused Courses
This type of course focuses on security principles and
techniques in building a software system. Network, also
considered as a system, might be part of the course, but not the
main focus. The focus is mainly on software systems in
general. Operating systems, programs, and web applications
are usually used as examples in these courses.
If an instructor wants to ask students to design and
implement a real system related to system security, there are
several choices. (a) If the instructor wants to let students learn
how to use cryptography in a real system, the Encrypted File
System Lab is a good choice. (2) If the instructor wants to let
students gain more insights on access control mechanisms, the
Role-Based Access Control Lab and Capability Lab are good
candidates. (3) If the instructor wants students to learn some of
the interesting ideas in improving system security, the Address
Space Layout Randomization Lab and the Set- RandomUID
Sandbox Lab are good candidates. Because all these labs
require modifying the underlying operating system kernel,
these are labs are meant to be carried out in the Minix
operating system. These labs can be used as the final projects.
Networking-focused Courses
This type of a course focuses mainly on the security
principles and techniques in networking.
Programming-focused Courses
The goal of this type of course is to teach students the
secure programming principles when implementing a software
system. Most instructors will cover a variety of software
vulnerabilities in the course.
Software-Engineering-focused Courses
This type of a course focuses on the software engineering
principles for building secure software systems. For this type
of courses, all the vulnerabilities labs can be used to
demonstrate how flaws in the design and implementation can
lead to security breaches. Moreover, to give students an
opportunity to apply the software engineering principles that
they have learned from the class, it is better to ask students to
build a reasonably sophisticated system, from designing,
implementation, to testing. Our design/implementation labs
can be used for this purpose.
C. Textbook Coverage
Most of instructors do choose a particular textbook in their
courses. There are several popular textbooks that are popular
among the computer security instructors. We show how SEED
labs can be used along with those textbooks; in particular, for
each of the textbooks, we have identified the chapters that can
use our SEED labs as their lab exercises to enhance student’s
learning of the subjects in those specific chapters. Our results
are summarized in Table II. We have picked the following four
textbooks in our studies:
Introduction to Computer Security, by Matt Bishop
(published by Addison-Wesley Professional in October
2004). We refer to this book as Bishop I.
Computer Security: Art and Science, by Matt Bishop
(published by Addison-Wesley Professional in
December 2002). We refer to this book as Bishop II.
Security in Computing (3rd Edition), by Charles P.
Pfleeger and Shari Lawrence Pfleeger (published by
Prentice Hall PTR in 2003). We refer this book as
Pfleeger.
Network Security: Private Communication in a Public
World (2nd Edition), by Charlie Kaufman, Radia
Perlman, and Mike Speciner (published by Prentice
Hall PTR in 2002). We refer this book as KPS.
IV. EVALUATION
With the goal of this project being enhancement of student
learning in security education via laboratory exercises, the
focus of the current assessment effort is on the efficiency and
effectiveness of the SEED labs. The efficiency of the lab
exercises is quantified by the following metrics: level of
difficulty, time spent on lab is worthwhile, and clarity of
instructions. The effectiveness of the lab exercises assesses
student learning and is quantified by the following metrics:
student level of interest, attainment of learning objective, and
value of the lab as part of the curriculum. Students were asked
to submit a short questionnaire after completing each lab based
on these metrics. Student responses were confidential and
completion of the surveys was on a voluntary basis. Among all
the SEED labs we have developed (about 25), 11 of them have
statistically sufficient number of survey responses (we have
total 735 survey responses). Therefore, the survey results only
from 11 individual laboratory exercises were analyzed.
Eight of the eleven exercises are categorized as
TABLE II
TEXTBOOK MAPPINGS
Numbers indicate chapter numbers.
Vulnerability and Attack Labs, with their overall learning
objective focused on students learning the principles of secure
design, programming, and testing. Three of the 11 exercises
are identified as Design/Implementation Labs, with their
overall learning objective focused on students applying
security principles in designing and implementing systems.
Survey data from 5 of the 11 labs were gathered over 3 years
of implementation (2007-2009), with Total N for these labs
ranging from 39- 77 student respondents. Four of the 11 labs
were included in the curriculum for 2 of the 3 years, with 3 of
these labs not included in 2009. Total N for these labs ranged
from 37-60 student respondents. During 2009, 2 new labs were
introduced, with Total N ranging from 11-15 student
respondents. Student survey data analysis across all 11
laboratory exercises is summarized below.
Lab instructions were clear: Approximately, three-
quarters or more of respondents (range from 93% to
73%) agreed or strongly agreed for 10 of the 11 labs.
60% of respondents indicated similarly for 1 lab.
The time I spent on the lab was worthwhile:
Approximately, three-quarters or more of respondents
(range from 94% to 74%) agreed or strongly agreed for
all 11 labs.
The lab was a valuable part of this course: Over 90% of
respondents (range from 99% to 91%) agreed or
strongly agreed for 10 of the 11 labs. 82% of
respondents indicated similarly for 1 lab.
I have attained the learning objective of the lab: Over
three-quarters of respondents (range from 95% to 78%)
agreed or strongly agreed for all 11 labs.
Student level of interest in the lab: Approximately,
three-quarters or more of respondents (range from 92%
to 74%) reported a high or very high interest level for
10 of the 11 labs. 66% of respondents indicated
similarly for 1 lab.
Level of difficulty of the lab: Over one-half of
respondents (range from 100% to 54%) reported a
somewhat difficult or very difficult level for 8 of the 11
labs. 46% - 32% of respondents indicated similarly for
3 labs.
To further explore these results, survey responses for these
items were analyzed by student gender, year of lab
administration, level of student preparation, student familiarity
with Unix, and level of lab difficulty. Due to page limitation,
we are unable to put all the evaluation results in this paper.
Detailed evaluation and diagrams can be found from our
project web site.
V. CONCLUSION
To help enhance student’s learning in computer security
education, we have developed a suite of hands-on lab
exercises. These labs can be conducted in an environment that
is very easy to build using the computing facilities that are
already available to students in most universities. Based on
this environment, instructors can select among 25 well-
developed laboratory exercises for their courses. These labs
cover a wide spectrum of security principles and can
accommodate a variety of security courses. We have
experimented with these labs in our own courses for the last
seven years. The results are quite promising. More than 15
other universities have also used our labs, and their feedbacks
are also quite positive.
We would like to disseminate the SEED labs to a larger
audience of instructors. All our labs are publicly available for
use under the open-source license. Instructors are welcome to
modify our lab description, if they want to tailor the lab
descriptions to fit their courses. The entire lab environment is
built into a pre-built virtual machine image, and can be
downloaded from our web page. We have dedicated set of
assistants, funded from the grant provided by NSF, for
providing help to instructors if they run into problems in using
our lab environment.
ACKNOWLEDGMENT
We thank the anonymous reviewers for their comments and
kind suggestions.
REFERENCES
[1] P. J. Denning. Great principles of computing. Communications of the
ACM, 46(11):15–20, November 2003.
[2] J. M. D. Hill, C. A. C. Jr., J. W. Humphries, and U. W. Pooch. Using an
isolated network laboratory to teach advanced networks and security. In
Proceedings of the 32nd SIGCSE Technical Symposium on Computer
Science Education, pages 36–40, Charlotte, NC, USA, February 2001.
[3] C. E. Irvine. Amplifying security education in the laboratory. In
Proceeding IFIP TC11 WC11. First World Conference on INFOSEC
Education, pages 139–146, Kista, Sweden, June 1999.
[4] C. E. Irvine, T. E. Levin, T. D. Nguyen, and G. W. Dinolt. The trusted
computing exemplar project. In Proceedings of the 2004 IEEE Systems
Man and Cybernetics Information Assurance Workshop, pages 109–
115, West Point, NY, June 2004.
[5] D. Kolb. Experiential learning: Experience as the source of learning
and development. Prentice Hall, Englewood Cliffs, NJ, 1984.
[6] W. G. Mitchener and A. Vahdat. A chat room assignment for teaching
network security. In Proceedings of the 32nd SIGCSE Technical
Symposium on Computer Science Education, pages 31–35,
Abstract—Recent advancements in medicine have enabled
doctors to achieve never-before-seen levels of efficacy in treating
traumatic battlefield injuries. While these achievements have had
the wonderful effect of reducing overall casualties, they have also
caused a dramatic increase in the number of soldiers who have
been rendered unfit for military service because of physical
injury. The result of this is that many veterans find themselves
making an unexpected career switch; despite the fact that many
of these individuals are otherwise fully capable of being
contributing members of society. For this reason, Auburn
University, in partnership with Mississippi State University and
Tuskegee University have created an outreach program to
provide vocational training in digital forensics to America’s
newly disabled veterans. It is believed that this training will
positively impact the career opportunities for injured soldiers.
Early results support this belief and indicate that the soldiers
believe they have benefited from it.
Index Terms— Digital forensics, veterans, computer security,
law enforcement
I. INTRODUCTION
DVANCES in modern medicine have dramatically
improved the odds of surviving a traumatic injury. These
advancements (along with evolved tactics and technology)
have resulted in a sharp decrease in the number of US fatalities
during the recent conflicts in Iraq and Afghanistan in
comparison to previous conflicts of similar size and duration.
These advancements, effective as they have been, are still
incapable of completely treating many different types of
traumatic injury encountered on the battle field. The result of
this is that wounds that would have been fatal 30 years ago are
now debilitating. In many cases of debilitating injury, the
injured soldiers have been declared medically unfit for
continued service. Many of these soldiers find themselves
with no definitive plan for the future; frequently the soldier
had planned to remain in the armed forces for several more
years.
At the same time that the conflicts overseas are producing a
pool of people in need of a new career, there continues to be a
documented need in the field of criminal justice and
investigation. In this field, the number of job openings is
expected to grow by 20+% between now and 2018 [3][4][5].
Given the longstanding pattern of retired service members
entering the law enforcement and investigative fields following
their discharge, Mississippi State University, Auburn
University, and Tuskegee University decided to create a
program that would attempt to solve two problems
simultaneously by providing recently discharged soldiers with
the skills necessary to tackle the shortage of digital
investigators.
The rest of this paper is organized as follows: in section 2
we describe the curriculum that is being taught to the soldiers.
In section 3 we present the results of an initial survey designed
to provide an initial glimpse into the efficacy of the training
that is being provided to the soldiers. Finally, in section 4 we
provide some concluding thoughts and describe the program’s
future.
II. OVERVIEW OF THE TRAINING
A. Training Overview
This training program is offered via face-to-face instruction
on-site at varying Army posts throughout the continental U.S.
The program is sponsored by the National Science Foundation,
which enables all costs, including instructors and equipment,
to be provided free-of-charge to the Army. Approximately 44
hours of total instruction are provided to the soldiers. Soldiers
are given instruction for four hours per day over a ten to
eleven day period. Limiting the daily instruction to four hours
enables soldiers to continue receiving rehabilitation for their
injuries, as well as, allowing them time to review the material
before the next day’s class. The class is limited to twenty
students in order to maintain a favorable student-to-teacher
ratio. No restrictions have been put in place to prevent
soldiers with certain injuries, such as traumatic brain injury
(TBI), from participating, provided they can operate a
computer.
B. Required Skills
The skills required for participation in the training are
relatively modest. Soldiers entering the training are expected
to be reasonably proficient with the Windows operating
system. In other words, soldiers are expected to be familiar
with basic concepts such as the keyboard, mouse, monitor, and
computer/tower. In terms of operating the computer, it is
expected that participants will be familiar with basic operating
skills such as left/right clicking of the mouse and typing. It is
Eric S. Ismand and John A. Hamilton, Jr. Department of Computer Science & Software Engineering
Auburn University Auburn, AL 36849
{eimsand, hamilton}@auburn.edu
A Digital Forensics Program to Retrain
America’s Veterans
A
also expected that participants will be familiar with the
Microsoft Windows operating system: using the Start button
and task bar; opening and closing programs; minimizing
programs; etc. Beyond these basic skills participants are not
required to have any additional qualifications for participation
in the course. The lack of prior access to the participants made
it impossible to use anything other than the soldiers’ self-
reported computer proficiency when deciding who was
suitable for the course.
C. Training Methodology
Students in the course are taught using a combined
lecture/laboratory format. For any given topic, the instructor
will lecture for approximately two to four hours. During the
lecture, the instructor will provide a basic overview, applicable
theory, as well as a walk-through of any tools that might be
used by an examiner. Following the lecture students engage in
a laboratory exercise that allows them to gain experience using
the tools and techniques that have just been described by the
instructor.
D. Curriculum
The course is designed with the expectation that the
students have little or no prior experience with information
technology or law enforcement. The goal of the course is to
provide participants with a working introduction to the field of
digital forensics. For students who are already considering the
field of digital forensics, this course is also designed to
provide them with a refresher on many of the topics that they
will be tested on if they should decide to pursue a professional
certification in the field. The course covers the basics in a
variety of disciplines before moving into more advanced
topics. Hands-on exercises are interspersed throughout the
lectures in order to give the students experience using the tools
and techniques they are learning.
The following is a list of basic topics that are covered by the
training program:
Basic Computer Usage
Soldiers are trained on the basics of operating Microsoft
Windows. Ostensibly these are skills that should have already
been present before enrolling in the course. They are covered
again, just to be sure that all of the participants have some
exposure to the operating system that will be used prior to
starting the hands-on activities.
Overview of Computer Hardware
Participants are given a brief overview of the different types
of hardware present within a computer, with particular interest
given to the overall functionality provided by each component.
For example, students are taught what the role of the CPU is,
and students are taught about the different types (volatile vs.
permanent) and forms (Cache, RAM, disk) that memory may
take.
Writing A Business Plan & Small Business Advertising
The belief, when creating the curriculum for this program,
was that the majority of graduates that seek employment as
digital forensic experts will work for some form of law
enforcement. It was recognized at the time, however, that
there may be individuals in the program that have no desire to
become part of law-enforcement; that might still wish to work
as digital forensics investigators. For this reason a module was
added to the program to teach transitioning soldiers how to
form their own business, how to write a business plan, some
basic accounting, and possible locations they should consider
advertising.
Introduction to Cybercrime
When creating the curriculum, it was believed that a
relatively low percentage of the participants would have prior
direct experience with digital forensics. Digital forensic
science differs from the forensic science prominently featured
in many TV shows and movies. It was believed that students
would not have been exposed to the practices and procedures
of digital forensics. For this reason, an overview of
cybercrime was added. This module details the types of
crimes that are encountered; a basic overview of the types of
evidence gathered is presented; as well as a basic overview of
the procedures and techniques used to analyze the evidence.
An introduction to writing warrants is also provided. Much of
the material featured in this module is repeated elsewhere in
the course; it was believed that this repetition would benefit
the student.
Introduction to Forensics Tools
This module provided an introduction into the tools used
during forensic analysis: write blockers, imaging devices, etc.
Overview of Digital Evidence Search and Seizure Procedures
This module provides a more in-depth examination of the
procedures that should be used when searching for and seizing
digital evidence. The students are provided with a concrete
timeline that tells them what actions should be taken, and the
order in which they should be performed.
Introduction to Imaging and Hashing
This module provides hands-on experience with the creation
of images, as well as experience with the creation of the image
hashes that can be used to verify the integrity of the evidence.
Overview of Digital Storage Techniques
A basic overview of the different types of storage that are
typically seen: magnetic, optical, and flash. This module also
provides a brief discussion of the logical organization of disks
as well as a discussion of slack space.
Introduction to Digital Evidence
A basic overview of digital evidence, describing when
digital evidence is permanently destroyed, when and where it
can be recovered, and other related topics. The subject of file
signatures and file carving is also introduced.
Introduction to File Systems
This module provides a beginner’s look at file systems:
what a file system is, how they are organized, as well as some
of the more important differences between the commonly
observed file systems.
E. Hardware and Software Tools
The students in this course are taught to perform their
imaging, hashing, and analysis using the Forensic Toolkit
software package by AccessData [1]. Students are trained
using hardware write-blockers manufactured by WiebeTech
[6].
During the course of training, students create images and
hashes of USB disks instead of traditional harddisks. The
reason for this is simple: time. The time savings of imaging a
1GB USB “thumb drive” as opposed to the smallest modern
hard disk is immense. Furthermore, the only difference from a
procedural standpoint, is the use of a USB writeblocker
instead of a SCSI/IDE/SATA writeblocker.
F. Course Scheduling
As mentioned earlier in this paper, this program was
purposefully structured to be taught over a two week period.
Because the targeted group of soldiers is typically undergoing
rehabilitation, it was decided to only hold class for four hours
per day. This leaves half the day for the soldiers to continue
their rehabilitation and maintain their other prescribed
appointments.
The other item considered when designing the program was
the current medical and mental status of the trainees. While all
of the soldiers that enter into the program possess their full
cognitive faculties, some of the wounded veterans have
experienced traumatic brain injury (TBI). Soldiers with TBI
frequently find it difficult to maintain concentration for long
periods of time [2]. By only holding class for four hours per
day, soldiers with TBI are better equipped to fully process the
day’s instruction, ensuring that they stay with the rest of the
group and do not fall behind on the material.
Because the daily lectures are designed to be given in four
hour blocks, we typically elect to conduct two different
sessions of the program at a time: a morning session and an
afternoon session. The two sessions are identical. In cases
where a soldier experiences a time conflict with his/her
normally scheduled class time, he/she can sit in on the other
session, space permitting. In some cases soldiers have
attended both of the day’s sessions in order to increase
repetition over the material. We encourage this whenever the
soldier is able and space is available.
Cost Description
As previously mentioned in this paper, this program is
funded by the National Science Foundation. This allows us to
provide the instruction at no cost to the Army or the soldiers
taking the course. This fact has greatly increased cooperation
from the Army in scheduling courses and has greatly enhanced
our overall ability to reach the target audience.
Initial Results and Performance
Our motivation behind this training is twofold: first, we are
driven by a desire to provide valuable training to America’s
injured soldiers. Secondly we want to improve our teaching
methodologies and techniques.
During the entire project, teaching efficacy is measured by
providing students with a post-course questionnaire that seeks
to determine how much the students have learned from the
course. The following is an overview of what the soldiers
were asked rate or indicate:
Previous IT/Computer skill
Previous law enforcement experience
Previous IT professional experience
How much they learned in the course
How likely they were to consider a career in digital
forensics, both before and after the course
The likelihood of recommending the course to another
soldier
The results that have been observed to date have been
positive. The following results are derived from an admittedly
small sample of 25 participants. It is believed that surveys
from future participants will confirm these findings:
On a scale of 1-5, with 1 being very little and 5 being
considerable, soldiers were asked to rate the amount
they had learned from the course. The average
response was 4.36.
On a scale of 1-5, with 1 being highly unlikely and 5
being highly likely, soldiers were asked to rate how
strongly they were considering a career in digital
forensics prior to taking the course. The average
response was 2.08.
On a scale of 1-5, with 1 being highly unlikely and 5
being highly likely, soldiers were asked to rate how
strongly they would consider a career in digital
forensics after taking the course. The average response
was 3.8.
On a scale of 1-5, with 1 being highly unlikely and 5
being highly likely, soldiers were asked whether they
would recommend the course to others. The average
response was 4.68.
Additional information about the participants:
The average age reported was 31.08 years
The average self-reported computer skill on a scale of
1-5, with 1 being very unskilled and 5 being very
skilled was 3.2.
Roughly 16% of respondents have reported working as
an IT worker at some point.
Roughly 16% of respondents have reported previous
law enforcement experience.
As shown in the data, we have seen a roughly 35% increase
in the number of soldiers giving some level of consideration to
a career in digital forensics. When asked to rate how much
they had learned and whether they would recommend the
course to a friend, soldiers rated 4.36 and 4.68 respectively.
This indicates that they find the training valuable and
interesting.
A detailed discussion of the demographics represented in
the course is limited by the data that was collected from the
participants. For example, while it is possible to state the
average age of the participants (31 years), it is not possible to
state what percentage of the participants was male or female.
This data and other items like it, such as race and educational
achievement, were not collected. The reason this data was not
collected was because it was not deemed to have a material
impact on the overall assessment of the quality of the training.
To-date, the partnership between our Universities has
resulted in approximately eighty students having completed the
program. Our partner University has trained a similar number.
We anticipate having trained between 300 to 400 servicemen
and women at the completion of the project between all the
partner institutions.
At the conclusion of the project we anticipate being able to
offer more additional details on the efficacy of this training
program. The number of graduates of this program that enter
the field of digital forensics is of particular interest.
Unfortunately this data is simply not available yet, as many of
the participants that have been trained to date have yet to be
completely discharged from their respective branches of
service.
III. LESSONS LEARNED
Though this study is only partially completed, our efforts to
date have resulted in several important “lessons learned”. First
and most importantly is a confirmation in our belief that a
training program of this kind was sorely needed. We have
been warmly and well received at every post we have offered
the training program at.
Not every discovery made during the offering of this course
has been a positive one. For example, scheduling courses has
been more difficult than originally anticipated. Our original
plan was to team with the Veterans Administration and offer it
through their hospital system. We devised this plan based on
the belief that they would be able to quickly and easily link us
with the soldiers we were trying to target. Unfortunately this
was not the case. The VA vocational rehabilitation program is
focused on veterans who are transitioning from active duty.
For this reason we have increasingly teamed with the “Warrior
Transition Units” based at most major posts. These groups,
which are organizationally part of the Army, have been very
supportive.
Anecdotal evidence indicates that our program is very
popular and highly in demand. Every time a course is held we
receive multiple requests for soldiers and people outside the
target group to be enrolled in the training program. Space
permitting we allow this, which probably helps to ingratiate us
to the host facility.
IV. CONCLUSION AND FUTURE WORK
As evidenced by the completion data, we are still in the
midst of executing this project. Thus far we have found the
reception from the Army and the soldiers to be quite positive;
many institutions have even asked us to consider making
return trips to their post to hold additional classes in order to
satisfy demand. The soldiers seem interested in the material
and report that they have learned a considerable amount from
this program.
The next logical step following completion of this work is
the creation of an online program. This will allow us to expand
our reach. At present the single greatest limitation to the
training we can provide are our physical limitations: number of
laboratory workstations, instructors, etc. Transitioning to an
online format eliminates many of these obstacles. There are
countless difficulties in attempting to provide forensics
training via the Internet, though, and these difficulties will
have to be overcome if meaningful online training is to be
made available to transitioning soldiers.
An alternate direction for our future efforts might be to
redesign the course with a focus on preparing participants for
one of the industry accepted certification exams. Obviously
this would be of great help to the students who believe they
wish to pursue a career in digital forensics, as it would
undoubtedly increase their likelihood of success on the
certification exam. We feel, though, that the basic
introductory course that we have developed is in better
keeping with our University’s outreach and service objectives.
Recall that the soldiers, sailors, and airmen targeted for this
training are individuals who are being medically discharged
from the armed forces against their will. These are people
who were not ready to leave the service and find themselves
making an unexpected career switch. It is believed that
offering these persons an introductory course at no-cost offers
them the opportunity to learn about a possible new field with
ample employment opportunities with no risk.
ACKNOWLEDGMENT
Our Lab Manager, Mr. James R. Thompson has played a
major role in this program. Mr. Thompson is an NSF
Scholarship for Service scholar and will join the US Civil
Service upon completion of his graduate studies at Auburn
University.
REFERENCES
[1] AccessData Inc. AccessData: A Pioneer in Digital Investigations Since
1987. February 15, 2010. http://www.accessdata.com/ (accessed
February 15, 2010).
[2] American Speech-Language-Hearing Assocation. "Traumatic Brain
Injury (TBI)." American Speech-Language-Hearing Association.
February 12, 2010.
http://www.asha.org/public/speech/disorders/TBI.htm (accessed
February 12, 2010).
[3] Federal Bureau of Labor and Statisitcs. "Private Detectives and
Investigators." Occupational Outlook Handbook, 2010-2011 Edition.
December 17, 2009. http://www.bls.gov/oco/ocos157.htm#outlook
(accessed February 13, 2010).
[4] Federal Bureau of Labor and Statistics. "Police and Detectives."
Occupational Outlook Handbook, 2010-2011 Edition. December 17,
2009. http://www.bls.gov/oco/ocos160.htm (accessed February 13,
2010).
[5] Federal Bureau of Labor Statistics. "Science Technicians." Occupational
Outlook Handbook, 2010-2011 Edition. December 17, 2009.
http://www.bls.gov/oco/ocos115.htm (accessed February 13, 2010).
[6] WiebeTech Inc. "USB Write Blocker." WeibeTech. February 15, 2010.
http://www.wiebetech.com/products/USB-WriteBlocker.php (accessed
February 15, 2010).
Roundtable Discussion: Forensic Education
Fabio R. Auffant II Technical Lieutenant,
Computer Crime Unit, NYS Police
Kevin Williams Chair, Psychology Department
University at Albany, SUNY
John A. Hamilton, Jr. Professor, CSSE
Auburn University
HE field of computer and digital forensics is changing rapidly and our dependence on computers and
network is increasing. Techniques from computer and digital forensics are being used not only for
investigating crime, but also for auditing systems as well as for recovery of lost data. Computer and
digital forensics involves data not only from computers, but also from servers, networks, and mobile
devices. The needs of the public sector workforce are growing as the demand for such expertise increases
within existing IT departments and new forensics divisions are created in agencies. However, they are
competing with the private sector, which often lure prospective employees with better salaries.
Knowledge of computer and digital forensics has become a necessary component of any IT specialist,
but due to the changing environment, it is also important to adapt by continuing to learn new tools and
techniques. Back by popular demand from last year, this round-table features a panel of experts who will
discuss the challenges faced by educators/trainers, law enforcement, and prosecution in terms of training,
retraining, and retaining a computer and digital forensics capable workforce. It will also cover novel
ways to ensure continuous training to security and forensics professionals. The panelists at the round-
table come from law enforcement, prosecution and academia and each brings their unique perspectives to
the discussion.
T
AUTHOR BIOGRAPHIES
Sabah Al-Fedaghi holds an MS and a PhD in computer science from Northwestern University,
Evanston, Illinois, and a BS in computer science from Arizona State University, Tempe. He has
published papers in journals and contributed to conferences on topics in database systems, natural
language processing, information systems, information privacy, information security, and information
ethics. He is an associate professor in the Computer Engineering Department, Kuwait University. He
previously worked as a programmer at the Kuwait Oil Company and headed the Electrical and
Computer Engineering Department (1991–1994) and the Computer Engineering Department (2000–
2007).
Victor Asal is an Associate Professor of Political Science at Rockefeller College at the University at
Albany, SUNY . Dr. Asal is a Research Associate with a focus on Ethnic Terrorism and Mass Casualty
Terrorism for the National Center for the Study of Terrorism and Responses to Terrorism (START). He
is also the Director of the Public Security Studies Certificate program at the University of Albany,
SUNY and is a co-director of the Project on Violent Conflict (PVC). Dr. Asal received his Ph.D. in
Government and Politics (Comparative Politics, Conflict and Conflict Resolution) from the University
of Maryland at College Park . He holds M.A.s in Government and Politics from the University of
Maryland at College Park and in International Relations (International Relations Theory, International
Law) from the Hebrew University of Jerusalem. Asal’s research has focused on terrorist behavior ,
including terrorist targeting of the United States structural and social antecedents to terror, non-state
actor mobilization on line and international crisis behavior. Much of Asal’s research efforts have
focused on identifying key factors from mostly qualitative literature, operationalizing these concepts in a
numerical fashion and then collecting the necessary data to quantitatively test the arguments in the
literature in a falsifiable fashion.
Fabio R. Auffant II has been employed by the New York State Police for the past 23 years. He is a
Technical Lieutenant in the Computer Crime Unit and Manager of the Computer Forensic Laboratory
located in the Forensic Investigation Center in Albany, NY, He has received extensive computer
forensics training, holds several certifications in Computer and Digital Forensics; he has conducted
forensic examinations and data analysis in hundreds of investigations, as well as testified extensively
throughout the State. T/Lt. Auffant is a member of several professional organizations such as, High
Technology Crime Investigation Association (HTCIA), International Association of Computer
Investigative Specialists (IACIS), Institute of Computer Forensic Professionals (ICFP), and InfraGard.
He has provided training and lectured to government and law enforcement personnel in the field of
Digital and Computer Forensics, as well as academic institutions such as John Jay College Criminal
Justice School, University of Albany Business School and NY Prosecutors Training Institute Summer
College. T/Lt. Auffant is an adjunct professor at Columbia Greene Community College in the Computer
Security/Forensics degree program and is the chairperson of the NY State Technical Working Group on
Digital & Multimedia Evidence.
Sriharsha Banavara is a graduate student at Auburn University in Auburn, AL pursuing a Master’s in
Software Engineering. He is currently seeking the degree with an Information Assurance option.
Alex Cowperthwaite is a masters student at Carleton University Ottawa, Canada, graduating in summer
2010. He received a bachelor of computer science from the University of New Brunswick in
Fredericton NB, Canada in 2008. Alex's research interests include network security, traffic clustering,
DNS Security and anomaly and intrusion detection.
John Crain is the Chief Technical Officer of ICANN. ICANN is responsible for coordinating domain
names, addresses, and other unique Internet identifiers. John has been contributing his technical
expertise and leadership to ICANN for over three years. In this role, he has taken on the task of
enhancing and improving the professional technical operations of ICANN and will lead the
organization's root management, website development and information services functions. John enjoys
an excellent and respected reputation with the global technical community. He conducts much of
ICANN's liaison work with the global technical community in order to facilitate and listen to discussion
on the many technical-based issues facing ICANN. John, a native of the United Kingdom and fluent in
Dutch and English, spent a significant portion of his professional career working in the Netherlands for
the RIPE NCC as part of the senior management team and was responsible for infrastructure, software
development and operations. Prior to RIPE NCC, John worked as a design engineer in research, design
and development of advanced materials.
John D’Arcy is an Assistant Professor in the Department of Management at the University of Notre
Dame. His research areas include information assurance and security and computer ethics. In recent
research, Dr. D’Arcy has examined the effectiveness of procedural and technical security controls in
deterring computer abuse. His studies also investigate individual and organizational factors that
contribute to end-user security behavior in the workplace. Dr. D’Arcy received a B.S. in Finance from
The Pennsylvania State University, an MBA from LaSalle University, and a PhD in Business
Administration with a concentration in Management Information Systems from Temple University. His
research has been published in journals such as Information Systems Research, Communications of the
ACM, Decision Support Systems, Computers & Security, and Human Resource Management.
Sanjukta Das is an Assistant Professor of Management Science and Systems at the State University of
New York, Buffalo. Her past research has been published in INFORMS Journal on Computing and
Journal of Organizational Computing and Electronic Commerce. She has presented her work at several
internationally recognized information systems conferences and workshops such as WITS, CIST and
WISE.
Wenliang Du is an associate professor of computer science in the department of EECS, Syracuse
University. He obtained his PhD in computer science from Purdue University. Dr. Du’s research
interests include security education, web security, privacy-preserving data mining, and security in
wireless sensor networks.
Noreen Gaubatz is the Assistant Director of the Office of Institutional Research and Assessment at
Syracuse University. She obtained her Ph.D. in Higher Education Administration from Syracuse
University. Dr. Gaubatz’s work supports a variety of university-wide assessment initiatives, with her
research interest in student ratings of teaching effectiveness.
Gwen Greene is a doctoral candidate in Applied Information Technology at Towson University. Her
research interests include risk management and information security. She is also an Information Systems
Security Engineer at Raytheon Company. Gwen has an undergraduate degree in Computer Science from
Bowie State University and a Masters in Information Systems from University of Maryland Baltimore
County. Contact her at [email protected].
Manish Gupta is a PhD candidate at State University of New York at Buffalo. He also works full-time
as an information security professional at a northeast US based bank. He received an MBA from SUNY-
Buffalo (USA) and a bachelor’s degree in mechanical engineering from I.E.T, Lucknow (India). He has
more than a decade of industry experience in information systems, policies and technologies. He has
published 3 books in the area of information security and assurance. He has published more than 50
research articles in leading journals, conference proceedings and books including DSS, ACM
Transactions, IEEE and JOEUC. He serves in editorial boards of 7 International Journals and has served
in program committees of several international conferences. He holds several professional designations
including CISSP, CISA, CISM, ISSPCS and PMP.
John A. “Drew” Hamilton is a professor of Computer Science and Software Engineering at Auburn
University and director of Auburn’s Information Assurance Center. He has a B. A. in Journalism from
Texas Tech University, an M.S. in Systems Management from the University of Southern California, an
M.S. in Computer Science from Vanderbilt University and a Ph. D. in Computer Science from Texas
A&M University.
Eric S Imsand, is an Assistant Research Professor of Computer Science and Software Engineering at
Auburn University. He has previously served as a Visiting Assistant Professor in the Department of
Computer Science at the University of Memphis and as a Principal Engineer of Software Engineering
and Security Technologies at MaxVision LLC. His research interests include biometric authentication
and identification, computer and network security, end-user security training, and simulation security.
He holds a bachelor's degree, master's degree, and Ph.D. degree, all from Auburn University.
Karthick Jayaraman is a PhD candidate at the department of EECS, Syracuse University. His advisors
are Dr. Wenliang Du and Dr. Steve J. Chapin. His research interests are security education, system, and
web security.
Yingying Kang is a candidate for PhD in Operations Research the Industrial & Systems Engineering
Department at the State University of New York at Buffalo. Her research interests include mathematical
programming, analytical simulation modeling, transport of Hazardous material, and network
optimization. Yingying also served as a technical support engineer at Parametric Technology
Corporation.
Varun Narayanan Kutty is a Master of Science student of the Industrial Engineering department,
majoring in Operations Research at the University at Buffalo, SUNY. He completed his Bachelor of
Technology in Electronics and Communication Engineering from National Institute of Technology,
Trichy, India. His research interests are in Risk Management practices using derivatives, Portfolio
Optimization, Revenue Management for online advertisements, IT strategy and Business Analytics.
Malvika Pimple is currently pursuing her M.S. in Computer Science from the University of North
Dakota.
Mark Scanlon is currently a research Ph. D. candidate in Computer Science at University College
Dublin (UCD), under the guidance of Prof. Mohand-Tahar Kechadi as part of the UCD Centre for
Cybercrime Investigation. He received his B.A. degree in Computer Science and Linguistics in 2006 and
his research M.Sc. in Computer Forensics and Cybercrime Investigation in 2009, both from UCD. His
research interests include Computer Forensics and Cybercrime Investigation, Forensic Verification, Data
Mining, Networking and Peer-to-Peer protocols.
Raj Sharman is an Associate Professor in the Management Science and Systems Department at SUNY,
Buffalo, NY. He received his B. Tech and M. Tech degree from IIT Bombay, India and his M.S degree
in Industrial Engineering and PhD in Computer Science from Louisiana State University. His research
streams include Information Assurance, and Disaster Response Management, Business Value of IT,
Decision Support Systems, Conceptual Modeling and Distributed Computing. His papers have been
published in a number of national and international journals. He is also the recipient of several grants
from the university as well as external agencies.
Anil Somayaji is an Associate Professor in the School of Computer Science at Carleton University in
Ottawa, Canada. He received a B.S. (1994) degree in mathematics from the Massachusetts Institute of
Technology and the Ph.D. (2002) degree in computer science from the University of New Mexico. He
has served as the program committee chair of the New Security Paradigms Workshop (2008 and 2009)
and has served on the program committees of ACM CCS, USENIX Security, and RAID, among others.
His research interests include computer security, operating systems, complex adaptive systems, and
artificial life.
R. Karl Rethemeyer is Associate Professor of Public Administration and Policy and Chair of the
Department of Public Administration and Policy at the Rockefeller College of Public Affairs and Policy,
University at Albany - SUNY. Rethemeyer’s primary research interest is in social networks, their impact
on social, political, and policy processes, and the methods used to study such networks. Dr.
Rethemeyer’s work spans two programs of research. The first focuses on the structure and operation of
collaborative and policy networks in the public sector. This work examines the challenges inherent in
the management of collaborative provision of public goods and services and the political ramifications
of engaging nonprofit and for-profit organizations in that effort. Most recently, Dr. Rethemeyer received
the Accenture Advances in Public Management Award for his research in this area. Dr. Rethemeyer’s
other program of research focuses on terrorism, terrorist organizations, terrorist networks, and counter-
insurgency/stabilization operations. Dr. Rethemeyer is co-director of the Project on Violent Conflict
(PVC), a research center focused on these topics. His Department of Homeland Security-funded work
focuses on how networks affect the use of various forms of terrorism (including suicide and CBRN
attacks), the lethality of terrorist organizations, the propensity of terrorist organizations to attack civilian
targets, and the propensity to choose or eschew lethal violence. Dr. Rethemeyer is also a co-investigator
on two other projects funded by the Office of Naval Research focused on the development of IEDs by
the Irish Republican Army and computational modeling of stabilization and counter-insurgency
operations.
Billy Rios is currently a security researcher for Google where he studies emerging security threats and
technologies. Before Google, Billy was a Program Manager at Microsoft where he helped secure
several high profile software projects including Internet Explorer. Prior to his roles at Google and
Microsoft, Billy was a penetration tester, making his living by outsmarting security teams, bypassing
security measures, and demonstrating the business risk of security exposures to executives and
organizational decision makers. Before his life as a penetration tester, Billy worked as an Information
Assurance Analyst for the Defense Information Systems Agency (DISA). While at DISA, Billy helped
protect Department of Defense (DoD) information systems by performing network intrusion detection,
vulnerability analysis, incident handling, and formal incident reporting on security related events
involving DoD information systems. Before attacking and defending information systems, Billy was an
active duty Officer in the United States Marine Corps. Billy has spoken at numerous security
conferences including: Blackhat briefings, Bluehat, RSA and DEFCON. Billy holds a Bachelors degree
in Business Administration, Master of Science degree in Information Systems, and a Master of Business
Administration.
Kevin Williams is Professor of Psychology and Dean of Graduate Studies at the University at Albany,
State University of New York. He has served as Chair of Psychology and Director of the PhD program
in Industrial/Organizational Psychology. His major areas of research are: (1) human motivation and
performance, where he studies the self-regulatory processes that guide goal strivings and goal revision
over time, (2) the psychology of blame, where his work explores the social-cognitive processes that
underlie the allocation of blame for accidents, and (3) the psychology of security, where he investigates
the psychological factors that support and compromise information security. His research appears in
such journals as Journal of Applied Psychology, Human Performance, Organizational Behavior and
Human Decision Processes, and Academy of Management Journal. He received his Ph.D. in
psychology from the University of South Carolina in 1984.
Yanjun Zuo is an assistant professor at the University of North Dakota, Grand Forks, USA. He received
a Ph.D. in Computer Science from the University of Arkansas, Fayetteville, USA in 2005. He also holds
a master degree in Computer Science from the University of Arkansas and a master degree in Business
Administration from the University of North Dakota, USA. His research interests include survivable
and self-healing systems, security in pervasive computing, database security, and information privacy
protection. He has published numerous articles in those areas.
INDEX OF AUTHORS
Al-Fedaghi, Sabah S. p. 10-18
Asal, Victor p. 9
Auffant II, Fabio R. p. 67
Banavara, Sriharsha p. 37-41
Cowperthwaite, Alex p. 2-8
Crain, John p. 1
D’Arcy, John p. 42-49
Das, Sunjakta p. 19-22
Dhillon, Gurpreet p. 50-55
Du, Wenliang p. 56-61
Gaubatz, Noreen B. p. 56-61
Greene, Gwen p. 42-49
Gupta, Manish p. 19-22
Hamilton, Jr., John A. p. 37-41
Hannaway, Alan p. 31-36
Ismand, Eric p. 37-41, 62-66
Jayaraman, Karthick p. 56-61
Kang, Yingying p. 19-22
Kutty, Varun N. p. 19-22
Kechadi, Mohand-Tahar p. 32-36
Lande, Suhas p. 23-30
Pimple, Malvika p. 23-30
Rios, Billy p. 31
Rethemeyer, R. Karl p. 9
Scanlon, Mark p. 32-36
Sharman, Raj p. 19-22
Somayaji, Anil p. 2-8
Talib, Yurita Abdul p. 50-55
Williams, Kevin p. 67
Zuo, Yunjun p. 23-30