Thanks Ben, that explains a lot. Is this documented somewhere that we can refer to?
> On May 28, 2014, at 19:57, Ben Pfaff <[email protected]> wrote: > >> On Wed, May 28, 2014 at 07:51:26PM +0200, Ricardo Schmidt wrote: >> We are running OpenFlow 1.3, OVS 2.1.2, and Ryu. Switch and controller are >> in the same machine with virtualized interfaces. >> >> We are running some stress tests on OVS to check the accuracy of the >> reported statistics from flow removed messages. >> >> In a first test we inject an average of 100 simultaneous flows per sec in >> the switch port, in the second an average of 250 flows/s and in the third >> an average of 500 flows/s. >> >> It seems that the more flows/s the more erratic the behavior of the >> counters. The total sum of reported bytes and packets at the end (summing >> all flow removed messages) is always less than the number of injected ones. >> This problem increases with the increasing number of flows/s. We do make >> sure that our flow mod always include the flag asking to the switch to send >> the flow removed upon entry expiration. >> >> In the third case, with 500 flows/s, not only the counters are off, but we >> also surprisingly see many flow removed messages in which counters have >> value of zero. That is, as if the total forwarded packets and bytes was >> zero. >> >> Does anyone has an idea of what might be happening? > > Counters only get updated about once a second. If your flows are > short, they might never get updated. _______________________________________________ discuss mailing list [email protected] http://openvswitch.org/mailman/listinfo/discuss
