Transport protocols and bulk file transfer

[stanislav shalunov <>]

Attending:
        Cesar Marcondes
        Stanislav Shalunov (scribe)
        Chester Ruszczyk
        Lisong Xu
        Ivan Beschastnikh

Injong and Steven said they couldn't attend.

Call started: 12:07PM

> * Preliminary report on Google-funded summer work.

Cesar did research on internal queuing on Linux and FreeBSD.
Timestamp a packet as it goes through the kernels (IP queue, UDP
queue, etc.)., do statistics, and understand the nature of internal
delay.  All the code and the results are on SourceForge:
<URL:http://heterovoip.sourceforge.net/>.  Stas: What about Windows?
Cesar: I'm going to work on that.  Track TSC in libpcap and thrulayd.
Naturally, no internal breakdown into IP and UDP queue.  Windows
allows driver development.  One problem is preserving correctness of
UDP checksum.

Ivan: has been working on implementing a transport protocol:
<URL:http://fb-fr-ccp.sourceforge.net/>.  Datagram-based protocol.
The application has functional control of datagram delivery; useful
for multimedia application.  Tasks this summer: finish specifying the
protocol (still working on congestion control); implementing a
user-level library that uses Steven Senger's API.  Can transfer data
now without congestion control.  Have working file transfer.  The
future: exact specifications for congestion control.  Use in some
application.

Chester: after SC2005, would like to try.

> * Fall Member Meeting: we have a slot reserved on Monday,
>   September 19, 2005, at 1:30PM.

Will have a telephone.

> * SC2005 demo plans.

Steven and Stas would like to include Ivan's library in Steven's
ImmSeg application before then.  Steven was going to start looking at
Ivan's code in the middle of September.  Ivan will resume work on
September 23 and will start adding real congestion control to his
library; believes it is realistic to have it there before November 14.

Chester plans his own demo, which is going to take quite some of his
time, making it difficult to spare any time to look at anything else.
Once SC2005 is over, would be interested in looking at Ivan's library
as well.  Will also send a pointer to the group on eVLBI so Ivan can
read up some.  Will also probably look at some of the Ivan's
documentation and code before SC2005 and perhaps provide feedback if
there's any.

* Additional:

Ivan to look at Cesar's work and understand it.  This is what would be
used to develop0 noise filtering for delay-based congestion control.

Chester: What about the IETF's recent transport protocols work?  How
does it relate to what we're doing?  Stas: It's great that the IETF is
ready to be a bit less incrementalist; the focus of the work here is
on high-performance transfers (hundreds of Mb/s and single Gb/s),
while the IETF is still focused on single Mb/s speeds, which makes for
some difference in approaches.  Stas plans to send something there in
perhaps two weeks to make them aware of the work here.

Cesar: Speaking of high-performance transfers, was reading some
discussion of TCP offload engines on Linux mailing lists.  Stas: Isn't
convinced that these attack the performance problem in the right
place; a throwback to early Arpanet days, in many ways, with largely
the same problems (lack of flexibility, difficult bugfixes, extra
layer between application code and the network, etc.).

* Action items:

  a. All (except Stas, Jeff, and Chet): Review the draft API Steven
     sent out (v.6.) and get comments to Steven.  [Ongoing.]

  b. Injong: track the COMNET call for papers and keep the group
     posted when a date has been published.  [Ongoing.]

  c. Injong: Run CFP for congestion control protocols past the
     transport list and send it to the general public (e.g.,
     e2e-interest).

  d. Gu: to talk to Bob about license.

  e. Injong: Pushing out the design document.

* Next call: September 19, 2005, 1:30PM US/Eastern (physical meeting
  in Philadelphia with telephone access).

* Call ended: 12:46

-- 
Stanislav Shalunov              http://www.internet2.edu/~shalunov/

This message is designed to be viewed with 0.06479891g of NaCl.