[ovs-discuss] Coverage counter subfacet_install_fail

Alex Wang alexw at nicira.com
Thu Jul 31 14:49:23 UTC 2014


Hey Han,


On Thu, Jul 31, 2014 at 1:33 AM, Han Zhou <zhouhan at gmail.com> wrote:

> Hi Alex,
>
> We had to upgrade to OVS2.1, and everything looks much better.
> The flow miss rate decreased from 20K/sec to 1.5K/sec, which means in
> OVS2.0 there was problem of creating datapath flows (indicated by the
> counter subfacet_install_fail), which then resulted in high miss rate
> and high OVS cpu.
>
>

Still, want to know the kind if traffic you used in your test.




> However, there are still occasional increase of upcall_queue_overflow,
> and the cpu load of handler threads are not high (< 10 %).
> upcall_queue_overflow      0.0/sec     0.000/sec        0.1339/sec
> total: 15671
>
>

This sample is outdated, since there is no queue overflow in the
last 5 sec nor last minute.  so you cannot tell, that the cpu utilization
when the queue overflow happens.



> This problem should go away when dispatcher is removed, do you know
> when would it be in stable release?
>
>

The removal of dispatcher thread will help,



> Best regards,
> Han
>
> On Wed, Jul 30, 2014 at 6:11 PM, Han Zhou <zhouhan at gmail.com> wrote:
> > Hi Alex,
> >
> > In our environment this counter subfacet_install_fail increases
> > quickly about 100/sec.
> > Could you help tell what could be the reasons, and what would be the
> impact?
> > This is OVS2.0. (We would like to upgrade, but not this moment in that
> > production environment, so it would be good to understand the cause
> > and effect).
> >
> > Best regards,
> > Han
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20140731/1a55da54/attachment-0002.html>


More information about the discuss mailing list