MASSACHUSETTS INSTITUTE OF TECHNOLOGY HAYSTACK OBSERVATORY WESTFORD, MASSACHUSETTS 01886 Telephone: Fax: 978-692-4764 781-981-0590 24 June 2002 TO: Distribution FROM: Alan R. Whitney SUBJECT: 24 June 2002 e-VLBI telecon summary Attendees: Lee Foster, Pat Gary, Chuck Kodak, Kevin Kranacs – GSFC Steve Bernstein, Lorraine Prior, Peter Schultz – Lincoln Laboratory Tom Lehman – ISI-E Richard Crowley, Kevin Dudevoir, Hans Hinteregger, Arthur Niell, Mike Titus, Alan Whitney – Haystack Observatory This telecon is one of an ongoing series of telecons to prepare for gigabit/sec e-VLBI demonstrations between NASA GSFC and MIT Haystack Observatory using a combination of network facilities including all or part of Glownet, Bossnet, ISI-E, SuperNet, Max and GSFC/HECN. Status Reports The attached figures of the e-VLBI path have been updated to reflect current status and are pretty much self-explanatory; critical status items are indicated in red. In addition, the following comments are relevant: − Tom and Peter reported that there is still a problem with the OC-48 Bossnet service. It has temporarily been changed to GigE, which seems to work properly. A service person will be dispatched to try to diagnose and fix the problem in the next few days; the problem appears to be in an optical amplifier in the Newark, NJ area. − Routers at G10, H3 and H7 need configuration updates to support the e-VLBI path. Pat and Tom will consult with Jerry Sobieski and/or Dan Magorian to accomplish this. Bill Fink can assist with this task if Jerry/Dan are not able to do it. Goal is to have this work done by the end of this week. − Pat reported that, as a result of the creation of several VLAN’s for e-VLBI purposes, workstation L6 (pluto at GGAO) has had an IP change to 206.196.178.53. − Pat reported that another Summit 5i (designated K4A) has been added between K4 (Bldg 28) and L2 (GGAO). Additionally, a ‘one-armed’ Cisco GSR 12000/16 router is attached to K4A to connect the GGAO and Bldg 28 nets. Pat does not believe this additional equipment should affect net performance. All equipment in the e-VLBI path should support 9KB jumbo frames. minutes14.doc 6/24/2002 1 − G7B IP address has been changed to 140.173.174.22. Performance Testing After the conversion of Bossnet to GigE, Tom achieved >900 Mbps single stream TCP in both directions. Kevin D. conducted four 30-minute tests, achieving ~965 Mbps average from ISI to Haystack, and ~940 Mbps average from Haystack to ISI, based on 1-sec averaging using iperf. Kevin’s plots are attached (top two plots of each test are transmitter performance; bottom two plots are receiver performance). The bandwidth was measured using iperf-1.1.1 with 1 sec averaging. The cpu idle time was measured using vmstat with 1 sec averaging. No information on packet loss rates or retransmissions. The tests were done with the following parameters: mtu = 4470 Bytes mss = 4418 Bytes buf = 3976200 Bytes Quick summary of the results: 1] test 1: ISI(kame) -> Haystack(evlbihay) avg. bandwidth = 965991626 Mb/s, tx cpu idle time = 40% rx cpu idle time = 37% 2] test 2: Haystack(evlbihay) -> ISI(kame) avg. bandwidth = 934996557 Mb/s tx cpu idle time = 12% rx cpu idle time = 40% 3] test 3: ISI(kame) -> Haystack(evlbihay) avg. bandwidth = 969819409 Mb/s tx cpu idle time = 37% rx cpu idle time = 41% 4] test 4: Haystack(evlbihay) -> ISI(kame) avg. bandwidth = 951017228 Mb/s tx cpu idle time = 13% rx cpu idle time = 40% Kevin D. reported that application performance with two Mark 5 systems back-to-back is currently limited to ~450 Mbps due to an apparent bottleneck from memory to/from disk. He plans to further investigate, perhaps trying another motherboard with a faster CPU (current CPU is 1GHz P3) and PCI bus. Kevin D. reported that he had difficulty measuring round-trip ping time from ISI to Haystack and back to ISI; varies from 4msec (impossible) to ~14msec. Round-trip ping time from Haystack to ISI, however, is consistent at ~14.5msec. Reason for this inconsistency is unknown. Pat reported that Bill F. has done testing between GGAO (L6) and GSFC Bldg 28 (K6), achieving ~992 Mbps TCP single stream (likely in both directions) using nuttcp. Connection seems to be very solid. minutes14.doc 6/24/2002 2 Action Items Alan: Distribute updated workstation configuration information to group. Steve: Keep group updated on Bossnet status. Pat/Tom/Bill F.: Work with Jerry Sobieski/Dan Magorian to properly configure routers at G10, H3 and H7. To be done by end of the week (28 June 2002). Bill F: Distribute information on nuttcp routine. Bill F.: Set up reverse mapping so GGAO test workstation can be managed from Bldg 28. Next telecon Next telecon will be Thursday, 11 July 2002 at 2 pm. xc: Steve Bernstein, LL Jim Calvin, LL Lorraine Prior, LL Leslie Weiner, LL Herbert Durbeck, GSFC Lee Foster, GSFC Pat Gary, GSFC Chuck Kodak, GSFC Kevin Kranacs, GSFC Paul Lang, GSFC Aruna Muppalla, GSFC Bill Wildes, GSFC Dan Magorian, UMCP Tom Lehman, ISI Jerry Sobieski, Max Richard Crowley, Haystack Kevin Dudevoir, Haystack Hans Hinteregger, Haystack Arthur Niell, Haystack Joe Salah, Haystack minutes14.doc 6/24/2002 3 Westford Antenna Millstone RPE Bldg (Lorraine) Dedicated no scheduling GigE WS/ Mark 5 Summit 1i 1 A2 A1 GigE 8 Alpine A4 3804 A3 Dedicated no scheduling 28-1A 14 B1 Summit 15 WS/ Mark 5 100 Mb 28-1B B8 B4 B6 B7 Hub B10 B9 To HS switch B3 Corr Cntl Cmptr GlowNet Congress St Boston (~50 km) Amp Cisco 1500 Connection made for e-VLBI E3 GigE Alpine 3804 E2 E1 D2 Lincoln Lab (Lorraine/Steve/Jim) Schedule to be coordinated with LL E4A Test WS From ISI-E (’Superglide’) GigE E6 E4 GigE Test WS 3l (2 GigE + 1 OC-48) GlowNet (~100 km) D1 Bossnet (Steve/Jim) ISI-E Scheduling on Bossnet calendar E7 Juniper OC-48 1310 M20 E8 GigE E5 Cisco 1500 GigE B11 D3 WDM E5A WDM GigE 5i 13 (Shared with LL on scheduled basis) C3 16 GigE GigE Dell Poweredge 2500 GigE B5 B2 LL C2 Haystack Correlator WS/ Mark 5 Schedule to be coordinated with LL C1 MIT Campus WDM F1 E9 Other users (no control) To be upgraded to OC-48 155x Amp Amp F2 Bossnet currently running GigE pending investigation and repair Dell Poweredge 500SC (provided by Haystack) Figure 1: e-VLBI Path - Haystack to ISI-E evlbi011.cdr 24 June 02 ISI-E (Tom) Test WS Scheduling on Bossnet calendar GE Summit 5i Test WS G5 G7B G6 G7 G8 OC-48 WDM GE G6B Summit 5i G4 G6C Tuner Bossnet MAX @ ISI-E (Jerry) ponder G1 OC-48 GigE H2 GigE G3 G11 OC-48 Borrowed OC-48 interface - may not be able to keep G10 Needs configuration H4 No jumbo frames Juniper M160 GigE H8 GigE H9 J3 Test WS? Primary e-VLBI path Needs configuration J2 H3 Needs configuration H6 Other users ( no control) Cisco 4912 Switch GigE Summit 5i H10 GigE J5 J4 WDM GigE on loan from GSFC H11 2 GigE l’s LuxN J6 J1 Need details OC-48 Juniper M160 MAX/UMCP (Jerry/Dan) H7 No scheduling; sufficient capacity for e-VLBI Backup plan Juniper M40 G9 MAX performance node; need details Test WS? H5 H1 G12 GigE GigE Trans- G2 Switch LX (~14km) GSFC/Bldg 28 (Pat/Paul/Kevin/Bill F.) K4 K1 WDM Will try to schedule either K2 or K3 exclusively for e-VLBI experiment G4 Mac with Yellowdog Linux GigE K2 K4A Summit Summit 5i 5i K4B K3 K5A K4 Test WS Test WS K6 K6A GGAO/Bldg 201 (Pat/Chuck) GigE LX (~7.25km) 1310 K7 GGAO Antenna Trailer (Bill W./Chuck) Dedicated - no scheduling necessary (~500m) GigE Summit L1 5i K5B Cisco 12000/16 K6A Windows PC (Lee Foster) L2 GigE L3 L5 G4 Mac SX Mark 5 L4 Test WS L6 evlbi012.cdr Figure 2: e-VLBI Path - ISI-E to GSFC/GGAO 24 June 02