9 th November 2005David Foster CERN IT-CS 3 WAN Overview Connectivity: –T0-T1 This has the “real-time” issue of needing to get the data out from CERN as fast as possible. Dedicated infrastructure. –T1-T1 Firm models for T1-T1 data flows yet to emerge. Can share T0-T1 infrastructure but some additional dedicated infrastructure is expected. –T1-T2 Some thoughts that requirements could be high, especially if a 200TB, Tier-2 cache refresh is done frequently. (2 full days at 10Gbit/sec) Mixture of general purpose shared IP networks and dedicated infrastructure. The WAN infrastructure is growing “organically” with a central architecture for T0-T1 being defined by the T0-T1 working group. –No central funding. –Many independent activities. –Creating the opportunity for a large connectivity capacity
9 th November 2005David Foster CERN IT-CS 1 LHCC Review WAN
Status David Foster Networks and Communications Systems Group
Leader 9 th November 2005David Foster CERN IT-CS 2 WAN Overview,
Parties, Stakeholders Status of DANTE/GEANT Transatlantic
Status/Plans Contents 9 th November 2005David Foster CERN IT-CS 3
WAN Overview Connectivity: T0-T1 This has the real-time issue of
needing to get the data out from CERN as fast as possible.
Dedicated infrastructure. T1-T1 Firm models for T1-T1 data flows
yet to emerge. Can share T0-T1 infrastructure but some additional
dedicated infrastructure is expected. T1-T2 Some thoughts that
requirements could be high, especially if a 200TB, Tier-2 cache
refresh is done frequently. (2 full days at 10Gbit/sec) Mixture of
general purpose shared IP networks and dedicated infrastructure.
The WAN infrastructure is growing organically with a central
architecture for T0-T1 being defined by the T0-T1 working group. No
central funding. Many independent activities. Creating the
opportunity for a large connectivity capacity 9 th November
2005David Foster CERN IT-CS 4 T0-T1 Designed around the concept of
an optical private network. Dedicated bandwidth (10Gbit to each T1)
on top of which an IP network is built. End-End path involves CERN,
GEANT and NRENs in some cases and dedicated links in others. Should
provide very high, reliable performance. Provided by different
entities in different ways: Individual initiatives (e.g. Renater DF
to CERN) Funded programmes (e.g. DOE and USLHCNet) International
projects (e.g GEANT) 9 th November 2005David Foster CERN IT-CS 5
T0-T1 Networking Group Progress We have had a number of meetings
and agreed on an overall approach with many open questions. Some
results have been publishedAnd we have a wiki site for gathering
operational information.
https://uimon.cern.ch/twiki/bin/view/LHCOPN/WebHome As we move more
and more towards providing an production infrastructure needed for
service challenges in 2006, agreements on operational details are
being created. Meeting in Seattle 14 th november 9 th November
2005David Foster CERN IT-CS 6 9 th November 2005David Foster CERN
IT-CS 7 9 th November 2005David Foster CERN IT-CS 8 9 th November
2005David Foster CERN IT-CS 9 9 th November 2005David Foster CERN
IT-CS 10 9 th November 2005David Foster CERN IT-CS 11 CERN-US 2005
(Lambda Triangle) 9 th November 2005David Foster CERN IT-CS 12 GLIF
MAP 9 th November 2005David Foster CERN IT-CS 13 9 th November
2005David Foster CERN IT-CS 14 Tier-0 - Tier-1 Connectivity
Tier1LocationNRENsStatus dedicated link ASCCTaipei, TaiwanASnet,
SURFnet2.5Gbit to Netherlight, 1Gbit from Surfnet BNLUpton, NY,
USAESnet, LHCnet10Gbit to NY Pop, ESNet providing last-mile to BNL
CNAFBologna, ItalyGeant2, GARR10Gbit across Geant-2 Dec 2005
FNALBatavia, ILL, USAESnet, LHCnet10 Gb in production IN2P3Lyon,
FranceRenater10Gbit via private dark fiber Dec 2005
GridKaKarlsruhe, GermanyGeant2, DFN10Gbit across Geant-2 Dec 2005
SARAAmsterdam, NLGeant2, SURFnet2x10Gbit, will be replaced by
GEANT-2? NorduGridScandinaviaGeant2, Nordunet2006 PICBarcelona,
SpainGeant2, RedIris2006 RALDidcot, UKGeant2, Ukerna10Gbit to
Netherlight via UKLight. To be replaced by SJ5 and GN2 in 2006
TriumfVancouver, CanadaCanet, LHCnet10Gbit to Netherlight, transit
to CERN via Surfnet. Will carry mixed traffic initially 9 th
November 2005David Foster CERN IT-CS 15 Issues, Conclusions etc.
The GEANT-2 cost model for the additional 10G pipes is not clear.
It will be a full cost recovery, marginal cost model but the
precise figures are unknown. Ballpark figures are reasonable. There
are a number of circuits becoming available end 2005/early 2006 as
GN2 starts production. Total capacity exceeds the stated
requirement of 1.6GBytes/sec but it needs to, to enable catch up
and provide headroom. The role of the GLIF community in providing
additional connectivity (e.g. T1-T2) is not clear but is a very
interesting resource. There are no major T0-T1 networking technical
issues but the models for T1-T2 continue to evolve. A very large
aggregate capacity will be available for LHC. The challenge will be
to really utilise the networks as they come on-line.
