On 04/13/2017 07:30 PM, Bodireddy, Bhanuprakash wrote: >> On 04/13/2017 07:11 PM, Kevin Traynor wrote: >>> On 03/12/2017 05:33 PM, Bhanuprakash Bodireddy wrote: >>>> Conditional EMC insert patch gives the flexibility to configure the >>>> probability of flow insertion in to EMC. This also allows an option >>>> to entirely disable EMC by setting 'emc-insert-inv-prob=0' which can >>>> be useful at large number of parallel flows. >>>> >>>> This patch skips EMC lookup when EMC is disabled. This is useful to >>>> avoid wasting CPU cycles and also improve performance considerably. >>>> >>> >>> LGTM. How much does this improve performance? > > I found significant performance improvement when testing with few hundred > streams. I remember the improvement was ~800kpps with smaller packets. > This is for the reason that emc_lookup() invokes expensive memcmp() to > compare the netdev_flow_key in EMC and it takes up significant cycles. Longer > the 'key', worse the performance. >
I just tested and saw about the same improvement. The memcmp won't happen if emc-insert-inv-prob=0 but you would still check the EMC for the key, so it's nice to avoid that. thanks, Kevin. >> Ack for the series, >> Acked-by: Kevin Traynor <ktray...@redhat.com> > > Thanks kevin for the review and Acks. > > Regards, > Bhanuprakash. > _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev