Paxson presented a paper on calibrating clock measurements on sender
and receiver packet measurements. When doing timings on both sender
and receiver, seems important.
K.K. Ramakrishnan presented a paper on the effect of ack compression
on asymetric links (fast cable modem downlink and slow modem uplink).
Standard TCP got terrible utilization of cable modem. Presented some
solutions (what they call "bandwidth allocation").
Erich Nahem @ TJ Watson gave a WIP on minor hacks to TCP (and other
stuff) for speeding up web servers. One of the hacks was
piggy-backing FIN and exploiting delayed ack to redeuce the # of
exchanges for small web docs (sound familiar?), getting some 10-15%
latency improvement over fastether. Asked him if he tried any of this
out on a modem connection, but not yet. He gave me an oldish copy of
a TR on the work; I'll bring copies to the retreat.
Dave Nichol @ Dartmouth gave a WIP about the Scalable Simulation
Framework (SFF). Huge collaboration effort; some of you probably know
about it already, but this was the first I had heard of it. Claims
they can simulate a 750,000 node network on an SGI Origin2000. Wasn't
clear at all on what exactly was being simulated, but spoke of glueing
the x-kernel on top of the simulator this summer so it sounds like
they are headed down the path of simulating real protocol
implementations. When asked, said that they might have something
others can use by the end of the summer. In any case, sounds like a
good thing to look into on the simulator front.
Walby et al. @ UIUC had an extended abstract titled "Total
ACknowledgements: A Robust Feedback Mechanism fo End-To-End Congestion
Control". What they do in one sentence in their words: "Total ACKs
supplement cumulative ACKs by giving the number of packet segments
received beyond the next expected packet number."
-geoff