Title: DataTAG%20project%20Status%20
1DataTAG project Status PerspectivesOlivier
MARTIN - CERN
- GNEW2004 workshop
- 15 March 2004, CERN, Geneva
2Presentation outline
- Project overview
- Testbed characteristics and evolution
- Major networking achievements
- Where are we?
- Lambda Grids
- Networking testbed requirements
- Acknowledgements
- Conclusions
3DataTAG Mission
TransAtlantic Grid
- EU ? US Grid network research
- High Performance Transport protocols
- Inter-domain QoS
- Advance bandwidth reservation
- EU ? US Grid Interoperability
- Sister project to EU DataGRID
4Project partners
http//www.datatag.org
5Funding agencies
Cooperating Networks
6EU collaborators
- Brunel University
- CERN
- CLRC
- CNAF
- DANTE
- NIKHEF
- PPARC
- UvA
- University of Manchester
- University of Padova
- University of Milano
7US collaborators
- Northwestern University
-
- UIC
- University of Chicago
- University of Michigan
- SLAC
- Starlight
- ANL
- Caltech
- Fermilab
- FSU
- Globus
- Indiana
- Wisconsin
8Workplan
- WP1
- Establishment of a high performance
intercontinental Grid testbed (CERN) - WP2
- High performance networking (PPARC)
- WP3
- Bulk data transfer validations and application
performance monitoring (UvA) - WP4
- Interoperability between Grid domains (INFN)
- WP5 WP6
- Dissemination and project management (CERN)
9 DataTAG/WP4 framework and
relationships
HEP applications, Other experiments
Integration
HICB/HIJTB
Interoperability standardization
GLUE
10Testbed evolution
- The DataTAG testbed evolved from a simple 2.5
Gb/s Layer3 testbed (Sept. 2002) into an
extremely rich multi-vendor 10 Gb/s Layer2/Layer3
testbed (Sept. 2003) - Alcatel, Chiaro, Cisco, Juniper, PRocket
- Exclusive access to the testbed is granted
through an advance testbed reservation
application - Direct extensions to Amsterdam UvA/Surfnet (10G)
Lyon INRIA/VTHD (2.5G) - Layer 2 extension to INFN/CNAF over GEANT GARR
using Junipers CCC - Layer 2 extension to the OptiPuter project at
UCSD (University of California San Diego) through
Abilene and CENIC under way. - 1st L2/L3 Transatlantic testbed with native
10Gigabit Ethernet access.
11DataTAG testbed phase 1 (2.5Gbps)
VTHD/INRIA
Linux PCs
ONS15454
stm16 (FranceTelecom)
Linux PCs
ONS15454
STM64
SURF NET
SURFNET CESNET
r06gva Alcatel7770
stm64 (GC)
CNAF
GEANT
r06chi-Alcatel7770
Linux PCs
stm16(Colt) backupprojects
Alcatel 1670
Alcatel 1670
r05gva-JuniperM10
r05chi-JuniperM10
r04chi Cisco7609
stm16 (T-Systems)
s01chi Extreme S5i
Chicago
Geneva
1G ethernet
DataTAG testbed phase 2 (10Gbps) simplified
10G ethernet
2.5G STM16
10G STM64
Linux PCs
Linux PCs
10Gbps Optical wave (T-Systems)
ABILENE
GEANT
StarLight Force10
JuniperM10
Juniper T320
Juniper T320
Cisco7609
Cisco7606
VTHD/INRIA
StarLight Cisco6509
Alcate l7770
edoardo.martelli_at_cern.ch last update 20030909
12(No Transcript)
13DataTAG testbed
Alcatel Chiaro Cisco Juniper PRocket
14Main networking achievements (1)
- Internet landspeed records have been beaten one
after the other by the DataTAG project partners
and/or teams closely associated with DataTAG - Atlas Canada lightpath experiments during
iGRID2002 (Gigabit Ethernet) and Telecom World
2003 (10Gigabit Ethernet, aka WAN-PHY) - New Internet2 landspeed record (I2 LSR) by
Nikhef/Caltech team (SC2002) - FAST, GridDT, HS-TCP, Scalable TCP experiments
(DataTAG partners Caltech) - Intel 10GigE tests between CERN (Geneva) and SLAC
(Sunnyvale) (CERN, Caltech, Los Alamos Nationa
Laboratory, SLAC) - 2.38 Gbps sustained rate, single flow, 1TB in one
hour - I2 LSR awarded during Internet2 Spring member
meeting (April 2003)
15ATLAS Canada Lightpath trialsTRIUMF Vancouver
CERN Geneva through Amsterdam
CANARIE 2xGbE circuits
NetherLight
StarLight
SURFnet 2xGbE circuits
A full Terabyte of real data was transferred at
rates equivalent to a full CD (680MB) in under 8
seconds and a DVD in under 1 minute Wade Hong et
al 09/2002
Bringing effective data transfer rates below one
second per CD!
Subsequent 10GigE WAN-PHY Experiments during
Telecom World 2003
16On Feb. 27-28 2003, a terabyte of data was
transferred in 3700 seconds by S. Ravot of
Caltech between the Level3 PoP in Sunnyvale near
SLAC and CERN through the TeraGrid router at
StarLight from memory to memory with a single
TCP/IPv4 stream. This achievement translates to
an average rate of 2.38 Gbps (using large windows
and 9kB jumbo frames). This beat the former
record by a factor of 2.5 and used the 2.5Gb/s
link at 99 efficiency.
10GigE Data Transfer Trial
European Commission
Huge distributed effort, 10-15 highly skilled
people monopolized for several weeks!
1710G DataTAG testbed extension to Telecom World
2003 and Abilene/Cenic
On September 15, 2003, the DataTAG project was
the first transatlantic testbed offering direct
10GigE access using Junipers VPN layer2/10GigE
emulation.
Sponsors Cisco, HP, Intel, OPI (Genevas Office
for the Promotion of Industries Technologies),
Services Industriels de Geneve, Telehouse Europe,
T-Systems
18Main networking achievements (2)
- Latest IPv4 IPv6 I2LSR were awarded, live from
the Internet2 fall member meeting in
Indianapolis, to Caltech CERN during Telecom
World 2003 - May 6, 2003
- 987 Mb/s single TCP/IP v6 stream
- October 1, 2003
- 5.44 Gb/s single TCP/IP v4 stream between Geneva
and Chicago - 1.1TB in 26 minutes or one 680MB CD in 1 second
- More records have been established by Caltech
CERN since then - November 6, 2003
- 5.64 Gb/s single TCP/IP v4 stream between Geneva
and Los Angeles (CENIC PoP) across DataTAG and
Abilene. - November 11, 2003,
- 4 Gb/s single TCP/IP v6 stream between Geneva and
Phoenix (Arizona) through Los Angeles - February 24, 2004
- 6.25 Gb/s with 9 streams for 638 seconds, i.e.
half a terabyte transferred between CERN in
Geneva and the CENIC PoP in Los Angeles across
DataTAG and Abilene.
19Internet2 landspeed record history (IPv4IPv6)
Impact of a single multi-Gb/s flow on the Abilene
backbone
20Significance of I2LSRs to the Grid?
- Essential to establish the feasibility of
multi-Gigabit/second single stream IPv4 IPv6
data transfers - Over dedicated testbeds in a first phase
- Then across academic research backbones
- Last but not least across campus networks
- Disk to disk rather than memory to memory
- Study impact of high performance TCP over disk
servers - Next steps
- Above 6Gb/s expected soon between CERN and Los
Angeles (Caltech/CENIC PoP) across DataTAG
Abilene - Goal is to reach 10Gb/s with new PCI Express
buses - Study alternatives to standard TCP (Reno)
- Non-TCP transport (Tsunami, SABUL/VDT)
- HS-TCP, Scalable TCP, H-TCP, FAST, Grid-DT,
Wesley, etc
21Main networking achievements (3)
- Advance bandwidth reservation
- GARA extensions
- AAA extensions
22Where are we?
- The DataTAG project came up at exactly the right
time - Back in the late 2000, 2.5 Gb/s looked futuristic
- 10GigE, especially host interfaces, did not
really exist - However, it was already very clear that the
standard TCP stack (Reno/Newreno) was problematic - Much hope was placed on autotuning
(Web100/Net100) ECN/RED like solutions - Actual bit error rates of transatlantic circuits
were over-estimated - Much better shape than expected on
over-provisioned RD backbones such as Abilene,
Canarie, GEANT - For how long?
- One of the strongest proof made by DataTAG is the
extreme vulnerability of production RD backbones
in the presence of high performance flows (i.e.
10GigE or even less)
23Where are we (cont)?
- For many years the Wide Area Network has been the
bottlemeck, this is no longer the case in many
countries, thus making the deployment of data
intensive Grid infrastructure, in principle,
possible, e.g. - EGEE the DataGrid successor
- Recent I2LSR records show, for the first time
ever, that the network can be truly transparent
and that throughput is only limited by the end
hosts and/or campus network infrastructures. - Challenge shifted from getting adequate bandwidth
to deploying adequate LANs and cybersecurity
infrastructure as well as making effective use of
it! - Non-trivial transport protocol issues still need
to be resolved - The only encouraging sign is that this is now
widely recognized - But we are still quite far from converging on a
practical solution?
24Layer1/2/3 networking (1)
- Conventional layer 3 technology is no longer
fashionable because of - High associated costs, e.g. 200/300 KUSD for a
10G router interfaces - Implied use of shared backbones
- The use of layer 1 or layer 2 technology is very
attractive because it helps to solve a number of
problems, e.g. - 1500 bytes Ethernet frame size (layer1)
- Protocol transparency (layer1 layer2)
- Minimum functionality hence, in theory, much
lower costs (layer12)
25Layer1/2/3 networking (2)
- Lambda Grids are becoming very popular
- Pros
- circuit oriented model like the telephone
network, hence no need for complex transport
protocols - Lower equipment costs (i.e. in theory a
factor 2 or 3 per layer) - the concept of a dedicated end to end light path
is very elegant - Cons
- End to end still very loosely defined, i.e.
site to site, cluster to cluster or really host
to host - Higher circuit costs, Scalability, Additional
middleware to deal with circuit set up/tear down,
etc - Extending dynamic VLAN functionality to the
campus network is a potential nightmare!
26 Lambda Grids What does it mean?
- Clearly different things to different people,
hence the apparently easy consensus! - Conservatively, on demand site to site
connectivity - Where is the innovation?
- What does it solve in terms of transport
protocols? - Where are the savings?
- Less interfaces needed (customer) but more
standby/idle circuits needed (provider) - Economics from the service provider vs the
customer perspective? - Traditionally, switched services have been very
expensive, - Usage vs flat charge
- Break even, switches vs leased, few hours/day
- Why would this change?
- In case there are no savings, why bother?
- More advanced, cluster to cluster
- Implies even more active circuits in parallel
- Even more advanced, Host to Host
- All optical
- Is it realisitic?
27Networking testbed requirements
- Multi-vendor
- Unless a particular research group is
specifically interested by the behaviour of TCP
in the presence of out of order packets, running
high performance TCP tests across a Juniper M160
backbone is pretty useless. - IPv6 achievable performance vary widely between
different vendors - MPLS QoS implementations also veary widely
- Interoperability
- Dynamic
- Implies manpower money
- Partitionable
- Reservation application
- Reconfigurable
- Avoid manual recabling, implies Electronic or
Optical switch/patch panel - Extensible
- Extensions to other networks
- Implies collaboration
- Not limited to network equipment, must also
include high performance servers, high perf.
Disks NICs, - Coordination with other testbeds
28Acknowledments
- The project would not have accumulated so many
successes without the active participation of our
North American colleagues, in particular - Caltech/DoE
- University of Illinois/NSF
- iVDGL
- Starlight
- Internet2/Abilene
- Canarie
- and our European sponsors and colleagues as well,
in particular - European Unions IST program
- Dante/GEANT
- GARR
- Surfnet
- VTHD
- The GNEW2004 workshop is yet another example of
successful collaboration between Europe and USA