On Tue, Nov 20, 2007 at 10:08:31PM +0800, Herbert Xu ([EMAIL PROTECTED]) wrote:
> Of course we still have the problem with the option in general
> that Dave raised. That is this may cause the proliferation of
> TCP receiver behaviour that may be undesirable.
Yes, it results in bursts of traffic because of delayed acks accumulated
in sender's lro engine, but from the first point, if receiver is slow,
then it will slowly send acks and they will be slowly accumulated, thus
changing not only seq/ack numbers, but also timings, which is equal to
increasing length of the pipe between users. TCP is able to balance on
this edge. I'm sure it depends on workload, but heavy bulk transfers,
where only lro with and without ack agregation can win, are quite usual
on long pipes with high performance numbers.
Until it is tested, I doubt it is possible to say it is 100% good or
bad, so my proposal is to write the code, which is tunable from
userspace, turn it off and allow people to test the change.
--
Evgeniy Polyakov
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
