> On 1. jun. 2016, at 15.42, O'Reilly, Darragh <darragh.orei...@hpe.com> wrote:
>
>> Using open vswitches v.2.5.0 with virtualization on Linux/KVM and having
>> random issues with cross VMs talk and wondering about increasing drops seen
>> on
>> the switch it self on my hypervisor nodes when similar drops are not really
>> seen on any VM ports or the physical ports.
>
> Run top on the host, and look for vhost-* kernel threads at 100% CPU
They don’t, see this link
<https://www.dropbox.com/s/slzqjiwwnu6zc8b/ovs_perf_issue_top-on-HNs.png?dl=0>
showing top snap across 6 out of 7 of our HNs (HA proxy LB VM running in top
left HN - this VM congest most traffic in/out of the application), neither do
any HN nor VM seem more than lightly loaded (cpu < 10%) eg. see HA VM stats
last 24 hours here
<https://www.dropbox.com/s/yl2boroempqecnx/HA-VM_stats.png?dl=0>.
Appreciate any good vswitch performance analysis/monitoring/tuning points, TIA
/Steffen
_______________________________________________
discuss mailing list
discuss@openvswitch.org
http://openvswitch.org/mailman/listinfo/discuss
