> Date: Mon, 15 Feb 1999 09:39:18 PST
> From: Venkat Padmanabhan <padmanab@microsoft.com>
>
> Your 3-dupacks scheme indeed appears to be an interesting
> solution that is compatible with the existing TCP standards.
> However, it seems to me that in the case of a sufficiently long
> outage, this scheme would not prevent cwnd from shrinking down
> to 1 segment, ssthresh from halving, etc. (correct me if I am
> mistaken!).
You're right, but that was a conscious decision. In the
mobility case we were dealing with, I would argue that you
want TCP to go into slowstart to probe the congestion state
in the new cell. That cell is unknown territory.
With a link outage that does not involve motion, slowstart
may not be what you want but I think it bears discussion.
My concern is that the congestion state along the TCP path
may have changed for the worse during the outage. It's
important for TCP to behave conservatively or risk making
a bad situation worse for everyone.
We can argue about whether slowstart does a good job of
utilizing available bandwidth after a link outage, but
I think that getting TCP out of the long retransmission
backoff fixes the first-order problem.
--Ramon
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:20 EST