Slow file transfers between TIGO (CL) and JIVE (NL)

PTS Case 18

Background

JIVE was involved in a program to track the SMART-1 spacecraft in its orbit around the moon. This involved telescopes in South America (TIGO in Chile and Fortaleza in Brazil). Several test observations had already been done, and data had been transferred electronically, for express analysis. The results were a little disappointing. So far there had been only a few, short transfers of TIGO data. The first test was done with FTP, and was subject to a locally imposed 5Mbps rate-limitation – 1.6Mbps was achieved. The second test used the Mark5 file transfer utilities and a tuned TCP stack, and the 5 Mbps restriction (imposed by the TIGO provider) was relaxed. Even though the path bottleneck was (reportedly) 100Mbps, only 7Mpbs was achieved. A week later there was a third, FTP test, and this achieved approximately 1.4Mbps. In separate tests JIVE deduced that there was little congestion on the 5Mbps TIGO link, since they were able to send UDP test packets at a rate of up to 4.6Mbps before packet loss was seen.

Investigation

For convenience the PERT began by running iperf tests in the reverse direction of normal data flow (Europe to Chile rather than Chile to Europe). It was seen that TCP parameters were negotiated well, including window scaling and large windows. It was also noted that Selective Acknowledgements (SACK) had been deactivated on the TIGO machine, and it was recommended this be reactivated. This seemed to have a positive effect, as subsequent tests achieved about twice the throughput, (3Mbps compared with 1.5Mbps).

Outcome

The poor performance was almost certainly due to congestion, most probably somewhere on the UdeC or REUNA networks. Using Linux buffer auto-tuining TCP performance seemed satisfactory (if not 100% consistent), often reaching 45Mbps. As REUNA planned to double the UdeC access link (to 310Mbps) it was decided to close the case at this point. A new case cold be raised if problems recurred in the future.

See also PtsCase22.

– Main.TobyRodwell - 31 Jan 2007