On Tue, 2008-05-20 at 22:18 -0400, Talpey, Thomas wrote:
> >
> >Does it make a measurable difference?  If so then it seems like a good idea.
It does a noticeable difference when the required message rate is high.
In those cases you spare the CPU from the need to write to memory,
possibly saving cache misses. I saw differences for IPoIB in OFED where
we use selective signalling for the UD QP.

> 
> I'll be happy to try it, but I bet it'll be hard to measure the difference
> with a storage workload. It sounds like a bit of a micro-optimization
> at the HCA interface, avoiding a few DMA cycles?
> 
> I didn't see a URL so let me know if so.
> 

http://lists.openfabrics.org/pipermail/general/2008-January/045071.html

_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to