Harvey welcomed all. We need to nominate a new chairperson for the Digital Divide area. The next ICFA meeting is in Amsterdam on July 30 2002.
Sylvain described the testbed used to obtain high throughput from SLAC to the US. They had a 622Mbps link. The time to recover from a loss was about 6 minutes, so a single loss is disasterous on the WAN. The RTT was about 120ms to Chicago and 160ms to Caltech. Introduced a new parameter called responsiveness rho=(C.RTT^2)/(2*MSS) determines how quickly go back to using the network link at full capacity after experiencing a loss. C is the capacity of the link. Values for rho are 1 hr 32 mins for C=10Gbps, RTT=120ms MSS=1460Bytes, with jumbo (9KB) get down to 15 mins. As a result have patched Linux for GRID TCP. Increase window size by M segments each time an ACK is received. Also looking at a smaller window reduction after a loss. This is regarded as an alternative to multi-streams. Get 280Mbits/s with one stream. Some problem that iperf cannot support long streams. RElated work is Sally Floyd use current window size in AIMD parameter. There is a Fast-TCP project headed by S. Low of Caltech. Solves TCP equilibrium and stability problems, use end-to-end delay rather than loss as congestion measure, very high utilization (99% in theory).
Peering at 2.5Gbps with GEANT via SWITCH, CERN has1Gbps to SWITCH. . Had expected to be at 622Mbps to US from KPNQWest. CIXP has some new partners. Plan to have 2.5Gbps DataTag and 622Mbps TransAtalantic from CERN to Chicago/StarLight, thence to ESnet, also to STARTAP (via 2*OC12).
KPNQwest delivered 622Mbps circuit but late delivery of Cisco 7609 routrers on CERN side and a faulty (unstable circuit) so had to extend 2*155Mbps contract. Then KPNQWest bankcruptcy, cancelled contract procured commodity Internet access in Geneva, and awarded new contract to DT.
Teleglobe crash caused problems for International connections. See http://www.slac.stanford.edu/xorg/icfa/jul02/ilya-russia.
The A&R net in the UK is working well. They are looking to an alternative to Worldcom. International connectivity is now via GEANT to the US using Level 3, still at 2.5Gbps from London to GEANT.
http://www.slac.stanford.edu/xorg/icfa/jul02/denis-france.pptRenater 3 Gbps backbone to be deployed by Sept. Lyon will have 1Gbps to REnater in Lyon (now 50+100Mbps). Renater now has 2.5Gbps link to US from OParis but for commodity Internet only, no peering. For REsearch will have 155Mbps to Chicago STATAP & GEANT. Will test Lyon-SLAC route through Renater, GEANT & Abilene+ESnet = Lyon- Paris_Frankfurt-NY-SLAC hope to get 600Mbps. Working with fiber providers to loan a fiber from Lyon and CERN via highway companies. Problem with last mile in Lyon (to be solved with REnater3) Cost estimate for 15 yr loan (downpayment + per year) 800KE + last mile + equipment = 60-80KE/yr.
No change in Internet connectivity still 622Mbps to each of Internet and ESnet. No formal confirmation from ESnet about upgrade of ESnet to 2.5Gbps.
Review recommended work with ESnet to develop hi-speed connectivity to STARlight in next 6 months. Main drivers are D0 & CDF. Expected upgrade to 622Mbps (from 155Mbps) at end July, but sub-contractor (Metro Fiber Netwokr) filed for Chapter 11. MFN had offered was $19K/month for dark fiber to STARLight. Will work with ESnet, also looking at electric company.
Satellite dishes being set up, Georgia will be first.
Suggestions for the WG included using the World Bank definitions of regions. Within regions want to identify sub-regional characteristics. World maps with lego bars or colored circles indicating performance also would be useful.
They have established a group. It consists of Vladimir Korenkov (JINR, Dubna), Olivier Martin(CERN), & Harvey Newman (Caltech). Will focus on various areas mentioned at last meeting.
Ties in heavily to monitoring to make the information available to the users.Our input is to put performance into user requirements. Will want to avoid duplicating work elsewhere such as in the Grid community and the Internet E2Epi. Includes many components, LAN, WAN, disk, file system, OS, application etc. User wants to be able to push a button to identify that the network is working OK. So one problem is parsing the problem.
An excellent resource is "Scientific Collaboration & Networking in SE Europe" www.pdg.ac.uk. Includes Albania, Bosnia, Herzogovina, Croatia, Yugoslavia, Macedonia, Kosovo, Serbia, Montenegro. Neighbors: Bulgaria, Romania, Greece, Hungary and Turkey (all have HEP communities). Aid/money for roads infrastructure etc. Not much for networking.
CERN was involved in gathering information on networking for SE Europe. See www.terena.nl/compendium has some info on SE Europe. Telcoms monopolies have traditionally got in teh way, and there are still a lot of monopolies in SE Europe. EU candidate countries are required to get rid of Telecom monopolies. SEEREN proposal (going ahead) is to link SE Europe countries at medium speed to GEANT via Greece. Romania & Bulgaria are already GEANT members.