It's nice to hear from 'the master's voice'.
>>>>> " " == Olaf Kirch <okir@monad.swb.de> writes:
> As a rule of thumb, whenever the congestion avoidance algorithm
> gives you problems, either your network or server is
> overloaded, or the _max_ congestion window is too big. You
> shouldn't be seeing timeouts on a big scale during normal
> operations at all. If you do, and try to remedy the situation
> by increasing the network load even more (by bumping the min
> congestion window) you usually make things worse.
Agreed. I realised this myself over a good night's sleep, withdraw
entirely my remarks.
I also realised more clearly that much of the congestion problem lies
not in the RPC layer, but in the NFS layers which are not checking for
and waiting for RPC backlogs to clear.
Specifically when we allow queued asynchronous writes, a wait-queue on
which tasks can sleep while the xprt layer clears the backlogs seems
to me to be necessary. I'm currently implementing this in the
write-gathering backport.
BTW: Now that I have you online. Are you planning on continuing the
NFSv3 development for linux-2.3? I'd like to see that done, and will
try to do it myself in between other commitments, but if you had plans
I'd be happy to defer ;-)...
Cheers,
Trond
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/