Issue Summary

The customer is the European Very Long Baseline Interferometry (EVLBI) Network, a group of astronomers who correlate data from several European radiotelescopes to produce very high-resolution images... sort of building a huge virtual radiotelescope. The sites involved are:

• EvlbiDwingelooSite^? in Dwingeloo, NL, where data from the radiotelescopes are correlated

and the radiotelescopes at

• EvlbiWesterborkSite^? in Westerbork, NL
• EvlbiOnsalaSite^? in Onsala, SE
• EvlbiTorunSite^? in Torun, PL
• EvlbiJodrellBankSite^? near Cambridge, UK
• EvlbiBolognaSite, Bologna, IT
• EvlbiAreciboSite^?, Arecibo, PR

http://cemp1.switch.ch/pert/evlbi-mails.html

• OPS-05-014-EVLBI-iperf-resuls-4-Feb-05.zip: Zipped results of iperf tests conducted 4 Feb 2005
• evlbitest10-02-2005.zip: Zipped results of eVLBI tests, 10 Feb 05

PTS (PERT Ticket System) pointer

https://www.pert.geant2.net/pert/preTicketDetails.do?ticketId=2

All new actions on this case should be documented here.

PRI (Pilot Roundup Issue System) pointer

http://issues.pert.geant2.net:8080/pert/issue4

There is still older information in the Roundup tracker. As soon as all of this has been migrated, this pointer can be removed

Detailed Tests

SwitchJbJiveTestsMay2005

Synagon plots from 04-Feb-2005 tests

Nature of the Tests

At around 1300UT (1400CET) TCP iperf tests were started, between each site and JIVE (NL). This ran for 25 mins. Then there was a pause of about 5 seconds, followed by about 25 mins of iperf UDP traffic.

Notes from Toby Rodwell on the plots:

These graphs show data transfer speeds per flow, sampled at 5 second intervals. A few things to note -

1. Problems with Synagon (it is not really production-quality software) means that I have not caught all the transfers (and in particular there is not much Italian traffic recorded).
2. All the graphs are in the form ---<sequence-number>. Unless the graph name includes the word 'out' then the graph shows traffic entering the GEANT network. If the sequence number has an 'a' or a 'b' suffix' then it is outside the main sequence of events and represent a snapshot taken at 1341UT ('a') and 1351UT ('b').
3. To compare the graphs make a note of
   1. The start time of the recorder
   2. The beginning x-point of the graph (elapsed seconds)
   3. The end x-point of the graph (elapsed seconds)
   4. The number of samples in the period (grey slider at top of graph)

I haven't really had time to think about the result myself yet, but I note a couple of things

1. I should set up an aggregate counter for traffic leaving GEANT bound for JIVE
2. Comparing the -PL-NL- and -PL-NL-out- graphs suggests it may in fact be interesting to graph traffic both entering and leaving.
3. It is always worth having at least 2, or even 3, instances of Synagon running, to allow for bugs crashing Synagon.

EVLBI - I'd be grateful to know what was configured/seen at the end systems, in terms of what tests started when, at which site.

IT -> NL

PL -> NL (received by GEANT from PL)

PL -> NL (sent from GEANT towards NL)

SE -> NL

UK (Manchester) -> NL

NL -> SE

NL -> UK (Man)

-- SimonLeinen - 08 Feb 2005