Hi folks:
Jonathan Stone and I have been slowly working for a good bit of this
year
on the checksum failure problem that Vern Paxson identified. In brief,
the problem is that about 1 in 7,000 datagrams has a bad transport
checksum
value and we are trying to understand why the failure rate is that high.
Unfortunately, due to lack of time, Jonathan and I have not been making
progress at the rate we would like to. The purpose of this note is to
see if there's someone in the E2E community (presumably, but not
necessarily,
a student) who'd like to help us out and get the basic work done.
We believe we have new packet capture software written that runs (though
the new person would have to check it to confirm we're right) and we're
close
to being able to run it on some systems which have large amounts of
traffic
running past them. We have also been offered access to a variety of
traffic dumps. That will give us a bunch of dumps to pore through
(perhaps
with new software required) to detect the types of errors occuring.
So expertise with C and some understanding of IP/TCP/UDP packet formats
is required.
Assuming we actually learn something interesting the new person would
co-author a journal paper with us about the results. (Warning: there's
no promise the results will be interesting -- there's still a
substantial
chance that the result will be "there's a bug in release 3.14159 of
So-and-So's kernel").
If this sounds of interest, drop me a note.
Thanks!
Craig
E-mail: craig@aland.bbn.com or craig@bbn.com