[ovs-discuss] Inaccurate statistics in flow_removed struct using OpenvSwitch
blp at nicira.com
Wed May 18 22:24:56 UTC 2011
It's busted again due to the merge from next. I have a patch out to
fix it again, though. (I haven't tested the accuracy, though.)
On Wed, May 18, 2011 at 03:22:51PM -0700, kk yap wrote:
> Thanks for the very fast response. TY tested the code and it worked.
> On 18 May 2011 12:55, Ethan Jackson <ethan at nicira.com> wrote:
> >> Looking at the code, ovs-openflowd returns facet + stat as packet
> >> count for the flow removed. ?The number recorded by facet is correct,
> >> but the sum of the value exceed the real value. ?This occurs at the
> >> facet_update_stats function. ?What is the purpose of the addition?
> >> Can we expect accurate statistics from OVS?
> > I should have responded to this question in my previous email. ?Packet
> > statistics are recorded in a best-effort fashion. ?In general they
> > should be accurate, but there is no hard guarantee that they are
> > especially in certain edge cases involving complex flow tables
> > including resubmit actions. ?In short, you should expect accurate
> > statistics from OVS noting the aforementioned caveat.
> > Thanks again for reporting this bug.
> > Ethan
> discuss mailing list
> discuss at openvswitch.org
More information about the discuss