Sagi Rotem wrote:
Do you get the same variance with the irqdaemon off ? How does the interrupts
on both cases spread between the cores ?
Under MSI-X, when I disable the irq daemon the mlx4 interrupts on each
node go to one CPU on each node.
I guess there might be an issue with the CPU affinity of the process /
network stack kernel thread and the CPU affinity of the hard irq handler
used by ipoib.
What's your recommendation regarding the affinity of those three elements?
Or.
_______________________________________________
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
