SWITCH to JB and JIVE Tests, May 2005

Subsequent to the measurements of Toby Rodwell, on 19/20 May 2005, I performed some BWCTL tests from a test box within the SWITCH network to the eVBLI sites Jodrell Bank (UK) and JIVE (NL).

BWCTL TCP and UDP tests carried out on 25 May with the following results:

Used hosts:
Jodrell Bank (UK): 192.84.79.27
JIVE (NL): 145.146.96.19
SWITCH (CH): 130.59.35.46

JB --> SWITCH: TCP: 360Mbps, 520Mbps UDP: 100/300Mbps loss 0.35%/2%
SWITCH --> JB: TCP: 600Mbps, 495Mbps UDP: 100/300Mbps loss -0%/-0%
JIVE ->SWITCH: TCP: 785Mbps, UDP: 100/300Mbps loss 0%/0%
SWITCH ->JIVE: TCP: 865Mbps, 890Mbps UDP: 100/300Mbps loss -0%/-0%

Remark: for some of the UDP tests iperf gives a -1 for the loss back. As I don't understamd the correct meaning of this negative value I regard this as no packet loss (this happens only if SWITCH is the sender of UDP packets).

These tests confirms the packet loss results from Toby (0.05% from JB to JIVE). In my opinion the primary PERT problem is the packet loss between JB and JIVE on the path JB to JIVE.

Correlating the lossy/lossless paths with the traceroutes below, the packet loss must have occurred somewhere in the path between:

 6  fr.uk1.uk.geant.net (62.40.96.90)  16.475 ms  16.456 ms  16.404 ms
 7  janet-gw.uk1.uk.geant.net (62.40.103.150)  16.357 ms  16.317 ms  16.272 ms
 8  po2-2.lond-scr3.ja.net (146.97.35.137)  16.432 ms  16.413 ms  16.296 ms
 9  po0-0.read-scr.ja.net (146.97.33.38)  17.728 ms  17.665 ms  17.565 ms
10  po3-0.warr-scr.ja.net (146.97.33.54)  21.377 ms  21.298 ms  21.272 ms
11  po1-0.manchester-bar.ja.net (146.97.35.166)  21.658 ms  21.679 ms  21.694 ms
12  gw-nnw.core.netnw.net.uk (146.97.40.202)  21.853 ms  21.893 ms  21.854 ms
13  192.84.79.27 (192.84.79.27)  22.505 ms  22.549 ms  22.548 ms

Suggestion:

Repeat BWCTL/iperf tests between all eVLBI sites and JIVE (both directions with TCP and UDP stream) to get updated performance results which reflect the current network and routes.

Localise the bottleneck (packet loss) on the path from JB to JIVE.

-- UlrichSchmid - 26 May 2005