[e2e] TCP un-friendly congestion control

[Sally Floyd]

>Even if we add "enough" reverse traffic, and
>short-lived web traffic with much diverse RTT ranges to remove the exact
>phase problems you are mentioning, we still see the same behavior where
>Scalable TCP does not converge under drop-tail or AQM with small
>buffers. I need to look further on HSTCP and its slow convergence to
>fairness and will let you know the results if you are interested.

Thanks, I would be interested in the HSTCP results.  In my simulations
(just started earlier this week, after a visit from Douglas Leith,
who is also interested in the convergence issue), HSTCP does converge,
but considerably more slowly with DropTail than with RED.  And I
haven't finished the simulations yet with the high-bandwidth links
(they take a longer time to run...)

In my simulations so far, I am assuming that the reverse-path TCP
traffic uses 10% of the link bandwidth, and that seems sufficient
to at least get rid of phase effects.  Though the DropTail simulations
are likely to have all of the perverse effects that one sees with
DropTail.  My current conjecture (not yet checked out) is that I
could control the convergence rate under DropTail simply by controlling
the charactertics of the reverse-path traffic...

- Sally
http://www.icir.org/floyd/