[e2e] trading acks...TRACKS
Jon Crowcroft
Jon.Crowcroft at cl.cam.ac.uk
Fri Nov 24 07:33:38 PST 2006
so a very simple scheme for rate controlling the internet
occurs to me based on the idea by christian and andy reported in
hotnets last year in:
"Using Packet Symmetry to Curtail Malicious Traffic"
Christian Kreibich, Andrew Warfield, Jon Crowcroft, Steven Hand, and Ian Pratt
(reminder: hotnets this year/next week in UC Irvine has some fine papers -
see the web already!)
so its this:
we don't have to send acks to the sender of a data packet - of course,
if we don't, then a well behaved TCP source will simply timeout and
rtx, although we could fix this if we sent a prior ack with zero
window
meanwhile, we _could_ send the same acks to someone else. i.e. acks
are "permits to travel" - with packet symmetry, of course, this is
even enforced...
how would we do this? well, if the "someone else's permit" field was
simply an IP address, this is simply a question of chaning a mapping
in a NAT box at two clients - i.e. we need NATs at two sites to swap
which global reachable addr they are using for whom and you're done
this could coordinate through many of the NAT traversal hacks that are
in superpeers (for example in skype and varius other apps way of
getting thru NATs) so for TCP based apps, one could control the
aggregate load towards a set of downloaders at the tracker/superpeer
sites....
for UDP sources doing TCP emulation at receiver, (TEAR) the same thing
would work (actually you could do it to RTCP reports, redirecting
those too) - leading to the TRACKS of the TEARs
hmmm....smokin
j.
More information about the end2end-interest
mailing list