Roland Dreier <[EMAIL PROTECTED]> wrote on 04/18/2006 03:01:57 PM:
> Shirley> After completion handler receives the notification, don't
> Shirley> poll the CQ right away, and wait for more WIKIs in
> Shirley> CQ. That way can reduce the CQ lock overhead.
>
> That's interesting... it makes sense, and it argues in favor of
> deferring CQ polling to a kernel thread. Of course this will hurt
> ping-pong latency. Maybe it's better to just implement NAPI though...
Let's try difference implementations to see the difference.
> Shirley> I found that there is some problem to increase the NUM_WC
> Shirley> on recv. It hurts the performance lots. Do you have any
> Shirley> clue?
>
> Is this on ehca or mthca? I couldn't explain it on mthca, but on ehca
> maybe there's a bug in generating CQ events ??
>
> - R.
It's on mthca. If you are interested. I can submit a test patch for your experimental.
Thanks
Shirley Ma
IBM Linux Technology Center
15300 SW Koll Parkway
Beaverton, OR 97006-6063
Phone(Fax): (503) 578-7638
_______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
