Transport protocols and bulk file transfer

[stanislav shalunov <>]

                          BULK TRANSPORT WG
                       Tuesday, April 25, 2006

Attendees:

Ivan Beschastnikh (by phone)
Larry Dunn (by phone)
Susan Evett (scribe)
Dan Hitchcock (ESnet)
Christian Rajagopalan (MPA)
Chester Ruszczyk
Steven Senger
Stanislav Shalunov 

Meeting started at 12:10PM Eastern time.

Agenda

Update on Ivan▓s project (VFER): Successfully ran a test from
Internet2 Ann Arbor office to San Diego.  Lots of framework for
delay-based congestion control, among other projects.  Delay-based
variation was delayed due to strange behaviors (delay increase not
seen until after packet loss occurs), queues being are overloaded too
fast, etc.  Current release uses loss-based congestion control (LBCC).
Plan is to have delay-based congestion control implemented and
documentation ideally before, and perhaps partially during, the Summer
of Code.

Stas reported on SoC 2006: More mentoring organizations, but cannot
gauge the interest in the project since submissions don▓t begin until
May 1.  Significant part of SoC at Internet2, is focusing on
contributions to the transport work.  Anyone can participate as long
as they are students.

http://transport.internet2.edu/soc2006/ideas.html

Stas reported that VFER is useful already and should become more so.
At this point, it is faster than TCP out-of-the box; Larry commented
that newest Linux, out-of-the-box, has auto-tuned send and receive
buffers, amongst other improvements.  Stas responded that the newest
release is improved, but it was still buffer-size limited in the
auto-tuning process.

Chet reported that, in about 2 months, he expects to have his
experiment (eVLBI) ready to test; Stas and Ivan reported that there
will be a much newer version of the code by that time.  Chet reported
that, when he implemented the framework, he would ensure that it was
easy to implement the new version without much effort.  He mentioned
an estimated 0.2% packet loss on the link where he's trying to test.
Dan responded that, in his experiments, about 1 packet/3 hours is the
highest level of tolerance for 10Gb/s transfers.

Next meeting to happen during the next Jt Techs workshop in July at
UW-Madison.

Call ended at 12:41PM.