Aaron,
> Having said all that, my understanding is that to get 'decent' performance out
> of TCP (ie., a window > 1) packet losses should be limited to less than 1 per
> RTT.
It had really ought to be darn better than that, since you're going to be
cutting your rate in half for every loss and building rate back at about 1
additional pkt per 2 RTTs. For any kind of reasonable (desired) bandwidth over
a long delay link, losses need to be pretty infrequent.
The one-loss-per-RTT that you cite was the magic number (pre SACK) that allows
you not to time out (arguably important, but not sufficient for "decent"
performance).
Regards,
Bob Durst
[email protected]
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:54 EST