Indeed, I need to be more careful to ensure my implementation patches match my test observations.
I'm not fully convinced of the need for a patch like this; My current plan is to tidy up the rest of the series, and do some further experiments before sending a separate follow-up for this. On 20 November 2013 14:55, Ben Pfaff <b...@nicira.com> wrote: > On Thu, Nov 14, 2013 at 03:28:29PM -0800, Joe Stringer wrote: >> This patch adds a new field to "struct ofproto" to track which protocols >> are active on that bridge. This is updated whenever bfd, cfm, lacp or >> stp is enabled. In instant_stats_run(), we query this to determine >> whether it is worthwhile to poll a port's status for each of these >> protocols. >> >> In a test environment running 5000 tunnel ports with only bfd running, >> we would previously see about 60% CPU usage when a port flapped twice >> per second. With this patch, average CPU usage decreases to around 55% >> in this case. >> >> Signed-off-by: Joe Stringer <joestrin...@nicira.com> > > A test like "if (status_mask | PSM_LACP)" is always true. It's almost > like writing "if (x + PSM_LACP)". Use & instead of |. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
