Hello!
> The following patch in will make the old behaviour come back. I assume
> this change was made to prevent over-scheduling
Yes.
> and thus the patch as it
> is might not be acceptable. Maybe a variant which tests against
> (sk->sndbuf - WATERMARK) could be?
It is difficult to believe that the problem is here, even if this
change kills the effect. Actually, I am even not sure, that this
is problem at all, congestion window is respected and burst
exist only when looking to tcpdump but not in realily.
TCP behaviour is controlled by ACKs, rather than by sndbuf.
BTW seems, you have too huge value for tx_queue_len on this link.
Set it to 4, it did not work earlier, but it should work in 2.3.
Alexey
|