OK - any suggestions on how to calculate the right value ? Could you explain the consequences of increasing this value ?
I ran ovs-vsctl set bridge xenbr0 other-config:flow-eviction-threshold=20000 the flows are stabilizing around 12-15000 Memory usage went from 12M to 14M - but the CPU load is still high > 94% But the latency actually went back down from 1.5 seconds to 500ms, which is normal. So this looks like it worked, even though the CPU usage remains high… Thanks. On 16/04/2013, at 21.52.46, Jesse Gross <je...@nicira.com> wrote: > On Tue, Apr 16, 2013 at 3:47 AM, Kristoffer Egefelt <kristof...@itoc.dk> > wrote: >> Thanks Jesse, >> >> I have a NAT firewall which logs something like this every second - OVS is >> processing ~400.000 packets/s - and I'm having issues with response times or >> even timeouts when load > 96%. >> Is there anything I can do about the latency/CPU usage, other than not >> running OVS on the NAT firewalls ? > > There is a fair amount of work going on to improve in this area. If > you have a large number of established flows (instead of just new flow > setups), one thing that might help is to use > ovs-vsctl set bridge BR other-config:flow-eviction-threshold=X > to set the threshold above your expected number of sustained flows. > The default value is 1000. > _______________________________________________ > discuss mailing list > discuss@openvswitch.org > http://openvswitch.org/mailman/listinfo/discuss > _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
