[e2e] Research update: CADPC/PTP
michael.welzl at uibk.ac.at
Tue Aug 26 08:02:33 PDT 2003
I would like to give you a brief research update.
Some of you might remember the Performance Transparency
Protocol (PTP); it is a generic means to efficiently
query routers along a path for performance related
information. I started discussing the protocol here a few
years ago, which was way too early.
In the meantime, I devised a congestion control mechanism
called "Congestion Avoidance with Distributed Proportional
Control" (CADPC), which solely relies on PTP feedback. It
works surprisingly well (given the small amount of feedback),
which I have shown using ns and published in:
Michael Welzl: "Scalable Performance Signalling and Congestion
Avoidance", Kluwer Academic Publishers, August 2003.
This is a revised (and the only actually published) version
of my Ph.D. thesis, which was supervised by Max Muehlhaeuser
and Jon Crowcroft - thanks!
Among other things, the mechanism converges to a max-min-fair
rate allocation in a fully distributed manner (i.e. without
ANY central logic or notion of number of flows). I believe
that I managed to show that PTP can indeed be useful.
CADPC is the result of a slow and gradual approach to reality,
starting from a very simple and unrealistic scenario. I made
a few rather severe assumptions about the environment, which
I intend to relax as the next step.
This next step may not be possible without funding - so,
if you work for a router vendor / ISP and have some money
to spare, I urge you to take a look at my PTP website :)
The PTP website has been updated with several new
information and, most importantly, thoroughly documented
and tested ns-2.26 code of CADPC and PTP. Please take
a look at it for further information:
PS: No, I'm not at SIGCOMM - couldn't make it this year -
but I wish y'all a good time in Karlsruhe!
More information about the end2end-interest