[e2e] Testing TCP-friendliness at routers

[Jon Crowcroft]

so one thing a router could do with TCP is to look at the sequence
number in packets - this would give you a distributed enqueuing
algorithm - in fact a stochastic fairqueueing algorithm could work as
follows

assuming at any given point in the net,  flows come from and go to
some fairly different prefixes, and the ISS selection of the TCP flows
is reasonably independnant, one could just hash the packets into the
queue position by tcp seqno%queusizer

an attack on this (cheat) would entail colluding between sender and
recipient and would prob. be reasonably hard - a check on the ISS (or
at any point in a flow on the seqno plausibility (e.g. rate it
advances comapred to other flows) ought to be easy to do cheaply
too....

sort of stochastic distributed fair queue insertion thru horrendous
layer violation...

In missive <1052900836.1772.83.camel at lap10-c703>, Michael Welzl typed:

 >>Hi all,
 >>
 >>Yet another question:
 >>
 >>The great floyd/fall paper "promoting the use of end-to-end
 >>congestion control in the internet" describes how to test
 >>for TCP-friendliness at a router in a fairly simple manner.
 >>Limitations of this test are also explained.
 >>
 >>Is this done in practice? Are there any results? Analyses?
 >>I suppose there should be some reports from experiments
 >>with this or similar techniques out there somewhere...
 >>
 >>Cheers,
 >>Michael
 >>

 cheers

   jon