For the record on this one, the researcher in question is
named Mark Smith, and he was, at least a year ago when I last
had contact with him, at MIT working with Nancy Lynch.
His home page, which contains pointers to some of his
work in this area can be found at <http://theory.lcs.mit.edu/~mass/>.
Quick summary for the impatient: T/TCP can get data corruption
and replay conditions under certain types of crash conditions
(A crash condition means a part of the network hickups. It might
be a machine at either end of the link, or it could be a router.)
Even stronger, unless all network transactions obey some
timing assumptions, no protocol that attempts to do what T/TCP does
can work. The open questions as I understand them at this point
are: (1) do the given assumptions get violated in realtity often
enough that we care? (for example, if the probability is signficicantly
lower than the probability of a data corruption that gets past the
checksum screen, then we probably don't), and (2) does T/TCP
as currently specified even work under the assumption that the
timing assumptions in question are obeyed.
Personally, I think T/TCP is a dead issue until these questions are
addressed in a complete and serious manner.
--
Eric Schenk www: http://www.loonie.net/~eschenk
email: eschenk@loonie.net, eschenk@rogers.wave.ca
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu