Post on 12-Jun-2020
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
© 2009 Jörg Ott 1
Towards a Delay-tolerant Future: Reconsidering some Assumptions in Networking
Jörg Ott <jo@comnet.tkk.fi>
2 July 2009 tubs.CITY
CHIANTI
© 2009 Jörg Ott 2
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Reconsidering the Internet Design…
Packets vs. circuits
QoS
Packet size (cell size)
End-to-end principle
Accounting
Number of layers
Layering principles
IP addresses
Locators vs. identifiers
IPv4 vs. IPv6 vs …
Security Trust
Best Effort
BGP
Routing structure Connectionless operation
MIddleboxes
Active networking
Hosts vs. interfaces
Communication paradigms
© 2009 Jörg Ott 3
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
The Internet
ISP
ISP
ISP
ISP
ISP
ISP
ISP
© 2009 Jörg Ott 4
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
The Mobile Internet is Not Connected What we have: lots of wireless network infrastructure
A patchwork of UMTS, WiMAX, WLAN, satellite (Iridium, …) covering significant parts of the planet… — yet not all
and powerful mobile endpoints with many of such radio interfaces To connect to the infrastructure and to talk to each other
What we still don’t have: sustainable ubiquitous connectivity for technical, practical, and economic reasons Also social and legal
Mimicking fixed network characteristics for wireless and mobile nodes is bound to fail
© 2009 Jörg Ott 5
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Example: Coverage
© 2009 Jörg Ott 6
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Performance Example: Throughput
© 2009 Jörg Ott 7
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Performance Example: Delay and Loss
Cellular data (UMTS, GRPS) in a train (2007)
Multi-access train system (2009) (WiMAX, UMTS, GPRS)
© 2009 Jörg Ott 8
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Link/path impairments/loss yield delay
Delay
Loss
Date Rate
“Short” delays inhibit interactivity and performance “Long” delays prevent service.
Disruption Data rate=0, loss=1, delay>Toutage
Reduced throughput,increased loss
Repair mechanisms, rate reduction
Del
ay
Delay Timeout Retry Failure “Manual repair”
© 2009 Jörg Ott 9
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Mobile Applications
Rely on (continuous) connectivity to the infrastructure
Essentially have not changed from the corresponding fixed applications.
© 2009 Jörg Ott 10
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Some Observations on (Mobile) Applications Many applications are asynchronous in nature
Theoretically no need for “always on” connectivity today Examples: e-mail, file transfer, peer-to-peer, even presence and messaging
Applications don’t communicate most of the time Users read, type, or do other things (e.g., fetch-read cycle for the web) Examples: web, e-mail, calendar, chat, presence, …
Application semantics do not require permanent or “end-to-end” connectivity… …but many application protocols do!
This motivates seamless connectivity and wireless QoS Which are often suboptimal goals and sometimes unrealistic
© 2009 Jörg Ott 11
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Scenario (0) Access Point
Web Content
Internet
© 2009 Jörg Ott 12
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Scenario (1) Access Points
© 2009 Jörg Ott 13
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Scenario (2)
Access Router
Access Points
© 2009 Jörg Ott 14
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Scenario (3)
Access Router
Access Points
DNS Server
Campus Network
© 2009 Jörg Ott 15
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Scenario (4)
Access Router
Access Points
DNS Server
Campus Network
Internet
Router
DNS Server
Router Router
© 2009 Jörg Ott 16
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Scenario (5)
Access Router
Access Points
DNS Server
Campus Network
Internet
Router
DNS Server
Web Server
Router Router
© 2009 Jörg Ott 17
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Infrastructure Dependencies Infrastructure functions
Autoconfiguration and access Address resolution User identification and authorization Security certificate lookup, validation Content retrieval (or other server-based operation)
Infrastructure interactions Autoconfiguration & access: 2 + timeout Address resolution: 1 per non-cached DNS name, possibly recursive
Web page: usually couple of external links Not so much applicable for mail servers, etc.
Applications Sending mail: ~6+ Retrieving mail: ~6+ Web: highly variable: 1 – 100+
spiegel.de 113 objects 590 KB tubscity 20 objects 343 KB tagesschau.de 106 objects 484 KB amazon.com 88 objects 692 KB wetter.de 118 objects 699 KB portal.acm.org 17 objects 77 KB
© 2009 Jörg Ott 18
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Infrastructure Dependencies
Information repository for retrieval and storage
Communication substrate
Communication mediation mechanism
A
S
A
B
S
S
A
B
© 2009 Jörg Ott 19
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Disconnections & Disruptions Delay Intentional disconnections and unexpected disruptions
Short- to long-term outage Reconnecting may lead to a change of IP addresses
Many short-term incremental fixes available from past research Mobile IP, HIP, VPNs to preserve IP addresses Transport layer mechanisms prevent losing TCP (and other) connections Session layer robustness prevents applications from noticing such (+ some support for selected applications)
BUT while disconnected, data cannot flow Connectivity loss, temporary path failures (re-routing) Hard to distinguish from congestion, remote node overload or failure Impossible to predict if and when operation will return to “normal”
Delays may have a variety of further origins…
© 2009 Jörg Ott 20
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Towards a Future Internet…
Three related issues to support mobile communications:
1. Communication substrate: Delay tolerance
2. Mediation: Enabling infrastructure-less operation
3. Revisiting application protocol design
© 2009 Jörg Ott 21
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
1) Towards Delay Tolerance State-of-the-Internet: deliver or drop
End-to-end approaches to resilience aim at suspending/resuming communications and at automating recovery
A
B
© 2009 Jörg Ott 22
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
1) Towards Delay Tolerance Connection splitting: prevent applications from noticing
Some issues Affects the end-to-end principle Introduces an additional (single) point of failure Need to identify challenged links Mobile nodes still require (direct) infrastructure access
A
B P
C P
© 2009 Jörg Ott 23
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
1) Towards Delay Tolerance From dedicated proxies to a routing overlay
Example DTN: Message-based communications Large messages instead of small packets Maintain the key packet forwarding of IP packets to preserve robustness
A
B
O
O
O
O
O
© 2009 Jörg Ott 24
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
1) Towards Delay Tolerance Eliminate the overlay:
Integrate delay-tolerant forwarding with the infrastructure nodes
No dedicated nodes for special handling required anymore
A
B
© 2009 Jörg Ott 25
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
2) Infrastructure-less Operation a) Retrieval and mediation indirection
Information-centric / content-based networking Decouple content and services from servers Address by intention (content, function, role) rather than server name
Reducing mapping/lookup indirection Route on contents, by intention Utilize late binding when addressing nodes Provide generic nodes for rendezvous support
Opportunistic and coordinated cooperation Simplest case: Caching Available to some limited extent as infrastructure-based overlays Extend this notion into access networks and to mobile devices
© 2009 Jörg Ott 26
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
2) Infrastructure-less Operation b) Autoconfiguration and registration
Provide persistent self-certifying identifiers for nodes Location-independent No need to dynamically obtain addresses for end-to-end operation (Lower layer addresses may stay for network layer interactions)
Litmus test: Applied to mobile (ad-hoc) environments Information exchange around a table Targeted communication and open sharing
© 2009 Jörg Ott 27
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
3) Fix Application Protocols Avoid chattiness: minimize end-to-end interactions
Keep messages self-contained (all-at-once) Indicate context, include credentials Minimizes the impact of delays
Untangle application and transport layer state Connections, reliability Allows repeated setup, support migration to more delay-tolerant transport
Separate protocol operation and content security Decouple content from the node (originally) serving it Keep application security independent of lower layers Enables caching and autonomous operation
Consider intermediaries explicitly in the design
© 2009 Jörg Ott 28
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
3) Fix Application Protocols Minimize the dependency on infrastructure nodes
Avoid third-party lookups and indirections Supports flexible content, service, and peer discovery
Design symmetric protocols that can talk directly without servers Realize end-to-end semantics expressly at the application layer Eliminates application layer conversion, e.g., in mail servers
Build adaptivity into protocols at all layers Allow users to define their own delay tolerance (rather than prescribing it) No fixed timeouts, flexible semantics, …
Non-delay-tolerant application protocols Can continue working as before (e.g., for real-time traffic)
© 2009 Jörg Ott 29
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Towards a Future Internet… Incremental deployment imperative
Don’t care about IPv4 vs. IPv6 vs. * No need to touch the infrastructure – but may evolve in parallel
Need to use TCP and UDP in the short to mid term
Realize delay tolerance and content orientation as an overlay Got plenty of real-world evidence for the latter today
Start out application-specific (plenty of evidence!) Overlay functions, content-aware routing, identifiers, … Gradually revise/enrich application protocols and applications (“dual stack”)
Strive for generalization, possibly using virtualization Intertwine overlay and underlay functions
© 2009 Jörg Ott 30
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Short-term example: The CHIANTI Approach
CHIANTI ISP
Internet
CHIANTI proxies
Servers, peers
3rd party provider CHIANTI proxies
Mobile Access ISP Internet
ISP A
ISP B
Access Mobile
© 2009 Jörg Ott 31
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Short-term example: The CHIANTI Approach Targets
Today’s applications: mail, web, … Incremental deployment Support for diverse business roles
Deployment considerations: NATs and firewalls Portals, authentication Content filtering
Mobile device
Vehicle Support Access Network Internet/CSP Server
U
C P CHIANTI overlay connection
Plain end-to-end connection
A B
A
B P C P
A
B O
O
O
O
O A
B
© 2009 Jörg Ott 32
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Mid-term example: Delay-tolerant Web Access
Message-based protocol: HTTP encapsulated in DTN Single request yields entire page Aggregation on the server side Compound responses in MHTML
Native DTN web server Existing web browsers + local
gateway Disconnected browsing +
blogging
Web client
web server
Challenged, disconnected, or regular network
R/O
R/O
R/O
R/O BR
HTTP-over-DTN with MHTML responses
Gate way
A B
A
B P C P
A
B O
O
O
O
O A
B
© 2009 Jörg Ott 33
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
Some Conclusion Disconnections, disruptions, delays are here to stay
Yes, we should advance the infrastructure and improve networks Delay tolerance should become an inherent property of a future Internet Support for wireless nodes, mobility, and robustness in general
Need for incremental deployment suggests overlays to start
Need to fix O(20 years) old application protocols
Prototyping and trialing is essential to understand the issues (usability, deployability, incentives)
HELSINKI UNIVERSITY OF TECHNOLOGY DEPARTMENT OF COMMUNICATIONS AND NETWORKING
© 2009 Jörg Ott 34
CHIANTI
http://www.netlab.tkk.fi/~jo/
http://www.chianti-ict.org/
http://www.future-internet.fi/