Introduction to National Supercomputer center in Tianjin TH-1A Supercomputer
ESnet and ANI Testbed Update - indico.bnl.gov...Provides world's largest collection (32) of...
Transcript of ESnet and ANI Testbed Update - indico.bnl.gov...Provides world's largest collection (32) of...
ESnet and ANI Testbed Update
Brian Tierney
DOE PI Meeting
Bethesda, MD
March 2, 2011
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Outline
1. ESnet Overview 2. ESnet R&D projects: OSCARS and perfSONAR 3. ANI Testbeds
2
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet is a Mission Organization
3
DOE Office of Science mission: to deliver scientific discoveries and major scientific tools for transforming our understanding of nature, and to advance the energy security, economic security, and national security of the United States.
ESnet mission: to accelerate scientific discovery for the DOE Office of Science by delivering unparalleled network infrastructure, services, tools, and innovation.
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet Supports DOE Office of Science
SC provides broad support for Labs, Facilities, people
• almost $5B/year in funding
• 45% of Federal support for physical sciences research
• key funding for basic research in biology, computing, energy, climate
• over 100 Nobel Prizes in past 60 years
Supporting > 27,000 PhDs, grad students, engineers at >300 institutions.
Provides world's largest collection (32) of scientific user facilities:
• supercomputer centers, accelerators, light sources, neutron sources, electron microscopes, nano-scale centers, a sequencing center, fusion facilities.
ESnet is one of them – connecting DOE sites, facilities, scientists and collaborators.
• optimized for science data transport
• every service exists to support scientific discovery
4
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet Supports DOE Office of Science
8
UniversitiesDOE laboratories
The Office of Science supports:27,000 Ph.D.s, graduate students, undergraduates, engineers, and technicians26,000 users of open-access facilities300 leading academic institutions17 DOE laboratories
SC Supports Research at More than 300 Institutions Across the U.S.
5
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet is Embedded in a Global Research Networking Ecosystem
3/2/12 6
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet is Embedded in a Global Research Networking Ecosystem
Most important implication: requirement for multi-domain coordination.
• Large-scale science is a multi-domain, end-to-end endeavor.
• We must build services across networks, exchanges, campuses. • Problem resolution, data mobility, measurement / monitoring,
guaranteed services, collaboration – all multi-domain!
• We spend much time coordinating, communicating • International: GLIF, DICE, LHCOPN, LHCONE, TERENA • US: Joint Techs, ESCC, Supercomputing, many more
Our drivers are different from those of commercial providers.
3/2/12 7
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet4 Topology (n x 10G core)
3/2/12 8
ESnet4 NetworkDrawn by Mark Redman
Revised:Revised by Mark Redman
!"#$%&'(%#$%)*+%,-.%/0$#12$$%3$2%-1456%,*%7*8%92:#$8+#;0826%<*+%=*(#2$%=*18'=8%#1)*>2$6128
US R&EInternationalR&E Peering
CommercialCommercial PeeringInternational Peering
OC12-Link(622Mb)NLR 10 GE Circuit
Internet 210GE-SDN-Circuit
Internet210GE-IP-Circuit
1GE-LinkSDN 10GE-Link
IP
MAN SITE
MAN 10 GE RINGSITE OC48-Link(2.48Gb) OC3-Link(155Mb) FE-Link(100Mb) DS3-Link(45Mb) T1-Link(1.45Mb)Ether-Link(10Mb)R&E and International
10G Peering ConnectionsR&E and International1G Peering ConnectionsPerfSonar enabled
perfSONAR4/14/2011
CANADA(CANARIE)
CENIC/DREN/INTERNET2/
LOS NETTOS/NASA/NLR/UW
ASIA-PACIFIC(KAREN/KREONET2/NUS-GP/ODN/REANNZ/SINET/TANET2/
TRANSPAC2)
AUSTRALIA(AARnet)
CENIC/INTERNET2/LOS NETTOS/NASA
ASIA-PACIFIC(KAREN/KREONET2/NUS-GP/
ODN/REANNZ/SINET/TRANSPAC2)
AUSTRALIA(AARNET)
LATIN AMERICA(AMPATH/CLARA)
EUROPE(GEANT)
MAN LANINTERNET2 / NLR /
NYSERNET
ASIA-PACIFIC(BINP/HEPNET)
CERNUSLHCNet
FRANCE(OPEN TRANSIT)
CANADA(CANARIE)
RUSSIA & CHINA(GLORIAD)
ASIA-PACIFIC(ASGC/KREONET2/TANET2)
CICNET-OMNIPOP/DREN/INTERNET2/MREN/NISN/NLR/NWU/STARLIGHT/UCHICAGO/
UIOWA/UWMADISON/WISCNET
SOX
FRGP
ABQ-GIGAPOPOSOGRANDE
TECHNET
MERIT/ULTRALIGHT
CERNUSLHCNet
CHINA(GLORIAD)
MAXGIGAPOP
NASA / NISNUS-DOI
3ROX/PSC
CANADA(CANARIE)
NOX
ASIA-PACIFIC(SINET)
INDIANA-GIGAPOPUI-ICCN
DREN/INTERNET2/
NASA
SANFORD
ONENET
LATIN AMERICA(CLARA / CUDI)
LATIN AMERICA(CLARA / CUDI)
INTERNET2
SDSC
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
perfSONAR
SNV-MR2
SDN1KANS
ATLA-CR1
CLEV-CR1
SDN1CLEV
BOST-CR1
SDN2SDSC
AOFA-CR2
ELPA-CR1
SDN1PNWG
SDN1ALBU
SDN1DENV
SDN1ATLA
SDN2CHIC
GA-RT2
PNWG-CR1
SDN1HOUS
HOUS-CR1
STAR-CR1SDN1STAR
SDN2WASH
BOIS-CR1
SDN2SUNN
SDN1ELPA
SDN1AOFA
SDN1NEWY
ALBU-CR1
DENV-CR2
NASH-CR1 SDN1NASH
ORNL-RT2
LASV-RT1WISCNET
KANS-CR1
SUNN-CR1 CHIC-CR1
WASH-CR1
CERN
EUROPE(GEANT/
NORDUNET)
SLA
C
SNLLLLNL
LBNL
JGI
NERSC
ANL
FNAL
EM
ORO
BJC
SRELSRS
LANL
KIR
TLA
ND
-AFB
DO
E-A
BLQ
INL
Y12
AMES
PU-HEP
GFDL
PPPL
NREL
IARC
GA
PAIX-PA
LIGO
PNWG
PNNL
MIT
JLAB
PNNL
PANTEXSNLA
ORNL
FORR
DOE-GTN
NNSA
NSO
EQX-SJ
EQX-ASH
BNL
EQX-CHI
ORAUNOAA OSTI
PNNL
SNL-DC
LLNL & LANL-DC
KCP
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet5 Optical Infrastructure Footprint
3/2/12 9
!"#$
%&!#
'#%&
(&)!
"*+"
'",-
'&.$
!$%&
),-)
"/01
01)0
!$#.
1&*!
-#%%
$*%!
2#,!
!#,#
3#.#
+&&-
"%'#
#%(*
-",4
'1&"'),"
%#!4
"01&
!#%$
"*."
.",&
(&!$#%(#
(*55
')$$
0%"4
!&*$
6#02
#$%#
01#$
'",!(#$&1&*%
623!
3"3'
,#!1
%&*)
0),0
01#.
.#%"
7#!1
'1)%
,"78
,"7+
#!1(
#,%
5,#%
&.,%
!,%%
,".!0
%(,%
6+)
!%#0
"!9:;<=&>;?@AB=,:;CDEF
4:EG?D9=<HI
JKLKK=JMNMOIJO
KI=%APQRAG
SJ=%APQRAG
)9T?9:EA=C?;U=JII+=%?;=VWII+=PAXY
#RR=-ED>=,DR:=V0?:9AY
&.,%="X>E:GG=,DR:=V0?:9AY
(,%
WW=%APQRAG
%)3#,=,DR:=V)9T?9:EAY
!*,,
!#0
Figure 1.
Figure 2.
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet5 L2/3 Architecture (n x 100G core)
3/2/12 10
!"#$
%&!#
'#%&
(&)!
"*+"
'",-
'&.$
!$%&
),-)
"/01
01)0
!$#.
1&*!
-#%%
$*%!
2#,!
!#,#
3#.#
+&&-
"%'#
#%(*
-",4
'1&"'),"
%#!4
"01&
!#%$
"*."
.",&
(&!$#%(#
(*55
')$$
0%"4
!&*$
6#02
#$%#
01#$
'",!(#$&1&*%
623!
3"3'
,#!1
%&*)
0),0
01#.
.#%"
7#!1
'1)%
,"78
,"7+
#!1(
#,%
5,#%
&.,%
!,%%
,".!0
%(,%
6+)
!%#0
"!9:;<=&>;?@AB=,:;CDEF
4:EG?D9=<HI
JKLKK=JMNMOIJO
KI=%APQRAG
SJ=%APQRAG
)9T?9:EA=C?;U=JII+=%?;=VWII+=PAXY
#RR=-ED>=,DR:=V0?:9AY
&.,%="X>E:GG=,DR:=V0?:9AY
(,%
WW=%APQRAG
%)3#,=,DR:=V)9T?9:EAY
!*,,
!#0
Figure 1.
Figure 2.
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
12,924 miles of Long-Haul Dark Fiber, ESnet5+
3/2/12 11 Figure 3.
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
One Important Driver for Innovation
3/2/12 12
1990 1995 2000 2005 2010
Total Yearly Traffic
Year
Traf
fic (P
B)
0.001
0.01
0.1
1.0
10.0
100.0
Although ESnet recently celebrated its 25th anniversary, its roots go back to the early 1980s, when two independent DOE-funded networks (HEPnet and MFEnet) merged. Those networks, which each served a single science – High Energy Physics and Magnetic Fusion – were on separate and incompatible technological paths. ESnet, designed to be protocol-stack agnostic, met the needs of both communities with a common infrastructure. In time, as IP became the lingua franca of the Internet, other protocols faded. Appendix (#), the ESnet Technical History, provides in-depth information on these early years.
Since its inception, ESnet has delivered services required by DOE science, many of which are cost-prohibitive (or simply unavailable) on the commercial market. Today, such services include
engineering, IPv6 service and performance parity, robust inter-domain multicast, continuous active performance monitoring, and
provides full operational transparency providing the community with valuable real-time information on network topology and utilization.
than the commercial Internet core, which sees millions of small
may last for days (as when data sets are transferred from the Large Hadron Collider [LHC] to Brookhaven and Fermi National Labs in the US). The price advantage ESnet customers enjoy was recently
under the stimulus-funded ANI project. Once this infrastructure is
deployed, the cost of adding new capacity will be a simple function of procuring new optical interfaces.
ESnet Production Services
In all its activities, ESnet strives to be at the leading edge of
solutions that advance the state of the art in distributed science. Leveraging its requirements workshops, collaboration with ESnet site technical coordinators and its peers around the world, ESnet has developed a portfolio of production services to meet users’ current and near future needs. The model of a consistent service set for science collaborations, delivered by independent but cooperating organizations worldwide, is a critical feature of ESnet’s research ecosystem and for which all of its services are engineered.
ESnet services include:
Core Services
» IP services
» Virtual circuits services
» Performance measurement and monitoring
Community Services
» Visualization tools
» Audio, video and desktop collaboration services
» Consulting support in data transfer, network architecture, and performance measurement
Core Services
IP Services
Network connectivity is oxygen for data-intensive science. A high-performance, feature-rich network is the foundation for
facilities. Routed IP is the most basic of Internet services, yet the effective support of science requires not only basic connectivity, but connectivity that enables high-performance applications. ESnet’s routed service is engineered to provide the highest level of support for demanding science applications, in particular high-performance TCP-based data transfers that are intolerant of packet loss.
ESnet has been an early adopter of new technologies, both to
applications. Two examples of this are IP multicast and the next-generation Internet protocol, IPv6. The DOE science programs used IP multicast for its collaboratories research, and ESnet was a key enabler of that research through its support of robust interdomain
allocation in North America, and the ESnet core network has
Perc
ent G
row
th
InternetESnet
Figure 2. CAGRPB Accepted per Compound Annual Growth Rates for ESnet [measured] and Internet core [estimated and predicted, see "Power Trends in Communication Networks", IEEE Journal of Selected Topics in Quantum Electronics, VOL. 17, NO. 2].
ESnet ESnet vs Internet
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet is Engineered for the Data Explosion Flood Tsunami Revolution
3/2/12 13
Cost of Sequencing per Genome
Date
Dolla
rs
Sep
2001
Mar
200
2Se
p 20
02M
ar 2
003
Oct
200
3Ja
n 20
04Ap
r 200
4Ju
l 200
4O
ct 2
004
Jan
2005
Apr 2
005
Jul 2
005
Oct
200
5Ja
n 20
06Ap
r 200
6Ju
l 200
6O
ct 2
006
Jan
2007
Apr 2
007
Jul 2
007
Oct
200
7Ja
n 20
08Ap
r 200
8Ju
l 200
8O
ct 2
008
Jan
2009
Apr 2
009
Jul 2
009
Oct
200
9Ja
n 20
10Ap
r 201
0Ju
l 201
0O
ct 2
010
Jan
2011
Apr 2
011
Jul 2
011
10,000
100,000
1,000,000
10,000,000
100,000,000
source: http://www.genome.gov/sequencingcosts/
Crushing data volumes producing qualitative changes in: • instrument design • workflows • collaborations • mission statements • cyberinfrastructure • discovery processes
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Elephant (or Alpha) Flows
3/2/12 14
Flow-Type Zoology, by Dimension: • size: elephant and mouse • rate: cheetah and snail • duration: tortoise and dragonfly • burstiness: porcupine and stingray
Kun-chan Lan and John Heidemann, “A Measurement Study of Correlations of Internet Flow Characteristics.” ACM Comput. Netw. 50, 1 (January 2006), 46-6.
Or: • Alpha flows (> H byes in < T sec) • caused by transfers of large files
over fast links
S. Sarvotham, R. Riedi, and R. Baraniuk, “Connection-Level Analysis and Modeling of Network Traffic,” ACM SIGCOMM Internet Measurement Workshop, November 2001, 99–104.
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Science Traffic is Different
3/2/12 15
and has already enabled researchers to make progress in multiple domains.
A few efforts that showcase ESnet’s successful transition of RDI concepts into production include:
typically had large bandwidth-delay products, and were extremely
dynamic virtual circuit services, which were deployed to supplement best-effort IP. The underlying technologies developed by ESnet — the OSCARS protocol and code base — have been adopted by over a dozen networks around the world including Internet2f in the US; RNP in Brazil; and NORDUnet, which serves the Nordic region of Europe. OSCARS also interoperates with similar services
f In addition, Internet2’s NSF-funded DyGIR and DYNES projects will in 2012 deploy over 60 more instances of OSCARS at university campuses and regional networks in the United States.
provided by GÉANT, the pan-European R&E network.
The latest release of OSCARS 0.6 was developed with a modular architecture that provides a consistent platform for researchers to experiment with more complex services, path computation and scheduling algorithms thus greatly easing the transition of their research from algorithm to a deployed service.
Understanding the health of the network is key for R&E networks like ESnet. A seemingly minor network fault, such as a dirty
performance. Performance issues can be challenging, especially
numerous organizations and time zones must collaborate on the debugging process. Poor performance not only hinders science productivity, but can lead collaborations to abandon networks and rely on physical shipment of disks or other media.
To address these issues, ESnet and several R&E partners have worked for over a decade to develop and standardize perfSONAR, which serves as an infrastructure for network performance monitoring, making it easier to solve end-to-end performance problems on paths crossing several networks. It contains a set of services delivering performance measurements in a federated
Tra!
c (T
B)
Figure 6.
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Science Traffic is Different
3/2/12 16
ESnet Accepted Traffic on January 28, 2012 (one day) http://www.es.net
http://www.ams-ix.net/statistics/
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
How do we Know what our Customers Need?
• Each Program Office at Office of Science has a dedicated requirements workshop devoted to its needs
• Two requirements workshops per year, with attendees chosen by science programs
• Discussion centered on science case studies
• Requirements derived from case studies + discussions
• Reports contain requirements, case study text
3/2/12 17
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet R&D
18
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet Research Success: OSCARS
OSCARS: an application and protocol suite for dynamic, inter-domain circuit management
• notable ESnet innovation
• used for service guarantees, traffic engineering, overlays
• carries almost ~50% of ESnet traffic (alpha flows)
• a customer-facing service: web interface for people, API for apps
• deployed in 21+ networks
3/2/12 19
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet Services: Core
OSCARS: an application and protocol suite for dynamic, inter-domain circuit management
• notable ESnet innovation • used for service guarantees,
traffic engineering, overlays • carries almost ~50% of
ESnet traffic (alpha flows) • a customer-facing service:
web interface for people, API for apps
• deployed in 21+ networks
2/1/12 40
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Production services OSCARS
Research Motivation • Network guarantees needed by service sensitive applications & large flows
Objective • Provide secure end-to-end guaranteed circuits across ESnet & multiple domains.
Impact • Carries 50% of ESnet’s 100+PB yearly data • Deployed in about 21 networks • Under evaluation by an additional 26 greenfield deployments
2003: Flow analysis and identification of large flows
2004: DOE funded
research project
2007: OSCARS production
service
2008: DICE collaboration
completes IDC protocol v1.0
2008: GLIF interops multiple implementations using FENIUS (ESnet driven)
2009: ARCHStone
research project funded
2011: Standards based NSI protocol defined in
OGF, successful interop with multiple
independent implementations
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
OSCARS Deployments Worldwide WAN, Regional, Campus, Testbeds
OSCARS
US: 13 Deployed 26 in progress
RNP
NORDUnet KOREN
JGN2
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet Research Success: perfSONAR
Distributed platform and protocol suite for performance measurement, monitoring, and visualization.
Assures end-to-end health of global networks.
400+ nodes deployed worldwide.
Developed in a global collaboration of R&E networking organizations.
Helps us find and fix problems!
3/2/12 22
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Network tools perfSONAR Research Motivation
• Certain performance problems only apparent to long-latency TCP flows, testing those requires source/sink in other domains
Objective • Provide an infrastructure that identifies and isolates performance problems in a
multi-domain environment.
Impact • Almost every ESnet site has fixed network performance problems • LHC collaboration has deployed extensively • 400+ perfSONAR hosts deployed on 100 networks and in 14 countries
2000: Formation
of OGF group
2003: First OGF doc on
network measurements
2005: non-standard GE performance
testers
2006: 1st release of software. perfSONAR
system 10GE b/w and GE OWAMP
2007:perfSONAR schema published
2008: 1st release of
perfSONAR-PS. 20 nodes
deployed
2009-11: Deployment at all 10G ESnet sites.
perfSONAR release every 6
months
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
perfSONAR in Use
Effect of full routing table, followed by reboot.
Effect of gradual optical line card failure.
1/29/2012 24
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Creating VO-Specific Dashboards from perfSONAR Data
1/29/2012
Joint Techs, Winter 2012 25
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
DOE’s Advanced Network Initiative (ANI) Testbed Project Update
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Testbed Value to Network Researchers
A realistic national-scale environment for innovative network research that would be impossible for most researchers to create in their labs
• Many types of network research require realistic high-latency environments
Maximum flexibility: researchers get “super-user” access to everything − ability to install custom OS on hosts − ability to do full routers/switch configuration − ability to install custom router software
A controlled environment that supports reproducible results
Priority is given to research that aligns with ESnet strategic focus areas, such as:
High performance network protocols, data transfer middleware, Openflow + MPLS integration strategies, energy efficient networking
3/2/12 27
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
ESnet Research Testbeds
Fertile environment for researchers to test and prove concepts
Control Plane Testbed • Long Island fiber, Infinera transport • Juniper, NEC OpenFlow,
Data transfer nodes, VMs
100G Testbed • High-speed protocol research • Routed network from Argonne to
NERSC
Dark Fiber Testbed • Disruptive network approaches • Quantum-key encryption
ANI 100G Router
nersc-diskpt-2
nersc-diskpt-3
nersc-diskpt-1
nersc-C2940 switch
4x10GE (MM)
4x 10GE (MM)
Site Router(nersc-mr2)
anl-mempt-2
anl-mempt-1
anl-app
nersc-app
NERSC ANL
Updated December 11, 2011
ANI Middleware Testbed
ANL Site Router
4x10GE (MM)
4x10GE (MM)
100G100G
1GE
1 GE
1 GE
1 GE
1GE
1 GE
1 GE1 GE
10G
10G
To ESnet
ANI 100G Router
4x10GE (MM)
100G 100G
ANI 100G Network
anl-mempt-1 NICs:2: 2x10G Myricom
anl-mempt-2 NICs:2: 2x10G Myricom
nersc-diskpt-1 NICs:2: 2x10G Myricom1: 4x10G HotLava
nersc-diskpt-2 NICs:1: 2x10G Myricom1: 2x10G Chelsio1: 6x10G HotLava
nersc-diskpt-3 NICs:1: 2x10G Myricom1: 2x10G Mellanox1: 6x10G HotLava
Note: ANI 100G routers and 100G wave available till summer 2012; Testbed resources after that subject funding availability.
nersc-asw1
anl-C2940 switch
1 GE
anl-asw1
1 GE
To ESnet
eth0
eth0
eth0
eth0
eth0
eth0
eth2-5
eth2-5
eth2-5
eth2-5
eth2-5
eth0
anl-mempt-3
4x10GE (MM)
eth2-5 eth0
1 GE
anl-mempt-3 NICs:1: 2x10G Myricom1: 2x10G Mellanox
4x10GE (MM)
10GE (MM)10GE (MM)
10GE (MM)
3x10GE (MM)
10GE (MM)
2x10GE (SM)[GT-TAM]
2x10GE (MM)
10GE (MM)
newy-tb-rt-1 (MX80)
newy-tb-of-1 (NEC switch)
bnl-tb-wdm-3newy-tb-wdm-1
bnl-tb-of-3 (NEC switch)
bnl-tb-rt-2 (MX80)
bnl-tb-wdm-4
3x10GE (MM)
1GE
1GE
bnl-diskpt-1 bnl-diskpt-2
newy-diskpt-1
bnl-app
1GE
1GE
1GE
2x10GE (SM)[GT-TAM]
newy-app
1GE
1GE
2x10GE (SM)
NEWY
BNL
AofA
100G Prototype Network
Updated July 8, 2011
2x10GE (SM)
bnl-mon
1GE
Long Island MAN (LIMAN) Testbed Architecture
2x10GE (SM)
10GE (MM)
2x10GE (MM)
1GE
1GE
(dashed line = planned)
ESnet Production Network
bnl-tb-of-2 (NEC switch)
10GE (MM)
!
Selma
Jackson
Houston
Dallas
Tulsa
Kansas CitySt. Louis
Peoria
Nashville
Louisville
Indianapolis
ChicagoSouth Bend
Cleveland
Buffalo
Albany
Cambridge
New York
Pittsburgh
Washington DCDenverGoodland
Albuquerque
El Paso
Los AngelesPhoenix
Echo Springs
Salt Lake CityReno
Sacramento
Sunnyvale
Eugene
Seattle
Boise
Raleigh
Philadelphia
Atlanta
Charlotte
Chattanooga
317 miles20
5 mile
s130 miles
152 miles 259 miles
264 miles
295 m
iles
179 miles
137 m
iles
228 miles255 miles
212
mile
s
150 miles 228 miles27
5 mile
s
212 m
iles
276
mile
s
116 miles
95 miles
560
mile
s
284 miles
618 miles
317 miles
422 miles
551 miles325 miles
500
mile
s
249 miles
863 miles
257
mile
s
278
mile
s
248 miles
172 miles
306 miles
275 miles
147 miles246 miles
198 miles 460 miles
336 miles
204 miles
532 miles
138 miles
LBNL Long Haul Dark Fiber Routes12,924 miles
BayExpres Metro Fibers: 432 milesChiExpress Metro Fibers: 167 milesNYExpress Metro Fibers: 6 miles
74 milesChepachet
Stamford61 miles
119 m
iles
Silver City
119 milesSeminary
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Community impact so far: Peer-reviewed Testbed projects
• 25 projects accepted: - 17: testbed proposal review - 8: direct DOE/ASCR funding - 5 from Industry; 8 from DOE
labs; 3 from NASA; 1 from DISA; 7 from Universities
• Strong diversity in research topics
• Four papers accepted or submitted for publishing, more to follow
3/2/12 29
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Control Plane Testbed
Control Plane Testbed
700 KM loop on Long Island Between Manhattan and Brookhaven National Lab
Infinera DWDM switches, NEC OpenFlow switches, Juniper router (soon with OpenFlow support), 6 hosts
Research Examples • OpenFlow experiments • Path Computation algorithms • MPLS/OpenFlow Integration
3/2/12 30
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Notes: - “App Host”: can be used for researcher application, control plane control software, etc. Can support up to 8 simultaneous VMs - “I/O Testers” are capable of 15 G disk-to-disk or 35G memory-to-memory - Other infrastructure not shown: VPN Server, file server (NFS, webdav, svn, etc.)
Control Plane Testbed, Long Island, NY
3/2/12 31
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
100G Testbed
100G Testbed
100Gbps path from Oakland, CA (NERSC) to Argonne, IL (ANL)
3 high-end hosts at each site, capable of >39 Gbps each
Research Examples • 100G data transfer tools − FTP100, GridFTP, BestMan, Saratoga
• TCP alternatives − RDMA, RoCE
• TCP enhancements − PERT, RAPID TCP
3/2/12 32
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Available as of Jan 3, 2012
3/2/12 33
ANI 100G Router
nersc-diskpt-2
nersc-diskpt-3
nersc-diskpt-1
nersc-C2940 switch
4x10GE (MM)
4x 10GE (MM)
Site Router(nersc-mr2)
anl-mempt-2
anl-mempt-1
anl-app
nersc-app
NERSC ANL
Updated March 1, 2012
ANI 100G Testbed
ANL Site Router
5x10GE (MM)
4x10GE (MM)
100G100G
1GE
1 GE
1 GE
1 GE
1GE
1 GE
1 GE1 GE
10G
10G
To ESnet
ANI 100G Router
4x10GE (MM)
100G 100G
ANI 100G Network
anl-mempt-1 NICs:2: 2x10G Myricom
anl-mempt-2 NICs:2: 2x10G Myricom
nersc-diskpt-1 NICs:2: 2x10G Myricom1: 4x10G HotLava
nersc-diskpt-2 NICs:2: 2x10G Myricom1: 4x10G HotLava
nersc-diskpt-3 NICs:2: 2x10G Myricom1: 2x10G Mellanox
Note: ANI 100G routers and 100G wave available till summer 2012; Testbed resources after that subject funding availability.
nersc-asw1
anl-C2940 switch
1 GE
anl-asw1
1 GE
To ESnet
eth0
eth0
eth0
eth0
eth0
eth0
eth2-5
eth2-5,20
eth2-5,20
eth2-5
eth2-5
eth0
anl-mempt-3
5x10GE (MM)
eth2-5 eth0
1 GE
anl-mempt-3 NICs:2: 2x10G Myricom1: 2x10G Mellanox
4x10GE (MM)
10GE (MM)
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Related Resources
Magellan at NERSC
currently 15 nodes with 1 10GE, GPFS backend
soon: 8 nodes with 2x10GE, GPFS backend
Magellan at ANL
went away in January
ORNL test cluster
12 nodes, each with flash-based scratch disk
FNAL test cluster
in planning stage
3/2/12 34
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Dark Fiber Testbed
Dark Fiber Testbed
Nationwide footprint (13,000 miles)
Researchers must pay to light the fiber, but no charge for access if results will be public
Research Examples • Greentouch Energy experiments • JPL quantum networking / quantum cryptography experiments
3/2/12 35
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
3/2/12 36
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Dark Fiber Network
10/18/11 5
Selma
Jackson
Houston
Dallas
Tulsa
Kansas CitySt. Louis
Peoria
Nashville
Louisville
Indianapolis
ChicagoSouth Bend
Cleveland
Buffalo
Albany
Cambridge
New York
Pittsburgh
Washington DCDenverGoodland
Albuquerque
El Paso
Los AngelesPhoenix
Echo Springs
Salt Lake CityReno
Sacramento
Sunnyvale
Eugene
Seattle
Boise
Raleigh
Philadelphia
Atlanta
Charlotte
Chattanooga
317 miles20
5 mile
s130 miles
152 miles 259 miles
264 miles
295 m
iles
179 miles
137 m
iles
228 miles
255 miles
212
mile
s
150 miles 228 miles27
5 mile
s
212
mile
s27
6 m
iles
116 miles
95 miles
560
mile
s
284 miles
618 miles
317 miles
422 miles
551 miles325 miles
500
mile
s
249 miles
863 miles
257
mile
s
278
mile
s
248 miles
172 miles
306 miles
275 miles
147 miles246 miles
198 miles 460 miles
336 miles
204 miles
532 miles
138 miles
LBNL Long Haul Dark Fiber Routes12,924 miles
BayExpres Metro Fibers: 432 milesChiExpress Metro Fibers: 167 milesNYExpress Metro Fibers: 6 miles
74 milesChepachet
Stamford61 miles
119 m
iles
Silver City
119 milesSeminary
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Current ANI Testbed Research
25 projects have been accepted to the testbed so far:
8 via direct DOE/ASCR funding
17 via testbed proposal review process
5 from Industry; 8 from DOE labs; 3 from NASA, 1 from DISA, 7 from Universities
Wide range of projects:
9: high-speed middleware
3: OpenFlow
4: other network control plane
1: 100Gbps end host hardware
1: network flow classification
http://www.es.net/RandD/advanced-networking-initiative/current-testbed-research/
3/2/12 37
1: quantum communications
2: Wide Area RDMA
2: TCP congestion control
1: security
1: energy efficiency
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Potential Breakthroughs from Current Testbed Research
Unconditionally secure continuous variable quantum communications
Ability to scale TCP well beyond 10Gbps
Alternative transport protocols that scale better than TCP (e.g.: RDMA)
Ability to easily create end-to-end circuits
Ability to scale data transfer tools/middleware to 100Gbps and beyond
100Gbps host interface
For more information see:
http://www.es.net/RandD/advanced-networking-initiative/current-testbed-research/
3/2/12 38
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Results to Date: Publications Submitted
After only 7 months of being operational, there were already several publications submitted based on ESnet’s Testbed results:
Energy Profiling of Network Elements for Rate Adaptation Technologies, M. Ricca, A. Francini, S. Fortune, T. Klein, submitted to the ACM/IEEE 3rd International Conference on Future Energy Systems (e-Energy 2012).
Traffic engineering in hybrid IP/optical circuit networks. Zhenzhen Yan, Chris Tracy, and Malathi Veeraraghavan, submitted to the IEEE 13th Conference on High Performance Switching and Routing (HPSR)
Middleware Support for RDMA-based Data Transfer in Cloud Computing, Yufei Ren, Tan Li, Dantong Yu, Shudong Jin, and Thomas Robertazzi, submitted to IEEE International Parallel & Distributed Processing Symposium (IPDPS)
Identifying gaps in Grid middleware on fast networks with the Advanced Network Initiative, G. Garzolglio and Dykstra, accepted at International Conference on Computing in High Energy and Nuclear Physics (CHEP 2012)
3/2/12 39
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Results to Date: Press Releases
Joint Press Release by Orange Silicon Valley, Bay Microsystems and ESnet on the first wide area 40G RDMA results (http://goo.gl/57gSJ)
• Showed that RDMA moves data at up to 96 percent of the peak capacity of the network − Much more efficient than TCP
• Replicated at SC11 over a 10,000KM path (http://goo.gl/ZE0w8)
Greentouch press release on plans to use ESnet dark fiber testbed for energy experiments: http://goo.gl/OB6nD
3/2/12 40
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Results to Date: Demonstrations
• ESnet Demonstration of OSCARS/OpenFlow integration on the ANI testbed at the Open Networking Summit held at Stanford. − http://goo.gl/4FFBg
• ESnet Demonstration of 9.9Gbps RDMA transfers from the ANI Testbed at BNL to Seattle at SC11
3/2/12 41
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Results to Date: Industry Involvement
• Loaner hardware for 40GE testing − Infinera, Alcatel-Lucent (ALU), Bay Microsystems, Mellanox
3/2/12 42
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
100G Testbed Performance
3/2/12 43
48.6ms RTT, 97.9Gbps aggregate TCP throughput* with 10 TCP streams
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Per-Stream Results
3/2/12 44
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Testbed Usage: Fully booked 24x7!
Feb 2012
March 2012
45
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Testbed Access
Proposal process to gain access described at:
https://sites.google.com/a/lbl.gov/ani-testbed/ Testbed is available to anyone:
• DOE researchers • Other government agencies • Industry
Must submit a short proposal to the testbed review committee • Committee is made up of members from the R&E community and
industry Goal is to accept roughly five proposals every 6 month review cycle
• Next round of proposals is due April 1, 2012
46
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Future Testbed Challenges
Long term operational support of the Testbed • Current testbed funding will run out in August of this year − Currently funded at 2FTE for testbed support
• This is essential for continued success
Hope to secure funding to ensure persistent/indefinite availability • 100G wave from NERSC to ANL will transition to production use
starting this summer, so need to identify funds to light a 2nd wave for testbed use
Dark Fiber will be available to the testbed until it is needed for production use
• For some segments this may not be for several years.
3/2/12 47
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
More Information
http://sites.google.com/a/lbl.gov/ani-testbed/
email: [email protected]