Shirley Ma wrote:
Rick,
Could you please run netperf/netserver on different CPU with the
irq handler to see any difference?
I have already done that - what is in the table is the peak performance out of
four different runs where I change the netperf/netserver CPU binding relative to
the interrupt CPU.
For which specific entries in the table would you like to see the four sets of
results? If you can give me the line from the table I can go back and find the
four results I ran to get there.
rick jones
BTW, what is "general-bounces" - it seems to have been one of the emails in the
dist...
The birectinal BW is much difference with the unidirection. We are
working on split CQ, send completion aggregation patch, and will test it
to see how much birectional BW improvement on Mellanox later.
Thanks
Shirley
_______________________________________________
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
