[ovs-discuss] Flow expiring before 5 seconds

[ananthan]

ohhh... thanks for that information i didnt know that,thanks  alot :)

but can i ask you one more foolish question relating this,

On one of my high traffic server when ovs-vswitchd reached 80 % CPU i saw
heavy latency 999ms for ping where usually it is 20ms ,and blamed
ovs-vswicthd for this as i thought it was its incapability to regenerate
expiring flow fast enough to handle icmp. So after  discussing with you it
looks like i need to look somewhere else to fix this, is that
so.Or because of many short flows packets are forwarded using slow path and
resulted in Latency.

using ovs 1.4.1
Thanking you,
Ananthan

On Thu, Jun 27, 2013 at 12:01 PM, Justin Pettit <jpettit at nicira.com> wrote:

> You can think of the datapath flow table as a cache of recently used
> flows.  Usually, ovs-vswitchd will keep a flow in the datapath for five
> seconds, but it may eject the flow sooner.  Under certain circumstances
> (e.g., lots of short lived flows), flows may not be set up at all.
>  However, regardless of whether flows are set up or not, the traffic should
> flow properly and not result in dropped packets.
>
> Did that clear things up?
>
> --Justin
>
>
> On Jun 26, 2013, at 11:25 PM, ananthan <ananthannair935 at gmail.com> wrote:
>
> > Can you please check now
> >
> > First Case:
> >
> > ping  -i .002 115.x.x.17
> >
> > watch -n1 "ovs-dpctl dump-flows xapi3 |grep icmp|grep 115.x.x.17"
> >
> >
> in_port(21),eth(src=1a:0e:aa:37:1e:6a,dst=00:00:0c:9f:f0:28),eth_type(0x0800),ipv4(src=5.x.x.10,dst=115.x.x.17,proto=1,tos=0,ttl=64,frag=no),icmp(type=0,code=0),
> packets:18323, bytes:1795654, used:-0.002s,
> actions:push_vlan(vid=40,pcp=0),1
> >
> >
> in_port(1),eth(src=00:26:0a:24:b0:00,dst=1a:0e:aa:37:1e:6a),eth_type(0x8100),vlan(vid=40,pcp=0),encap(eth_type(0x0800),ipv4(src=115.x.x.17,dst=5.x.x.10,proto=1,tos=0,ttl=62,frag=no),icmp(type=8,code=0)),
> packets:18324, bytes:1795752, used:-0.004s, actions:pop_vlan,21
> >
> > Flow didnt expire
> >
> > Ping Outcome:
> > --- 115.x.x.17 ping statistics ---
> > 29238 packets transmitted, 29185 received, 0% packet loss, time 85337ms
> >
> > Second case:
> >
> > ping -i 1 115.x.x.17
> >
> > watch -n1 "ovs-dpctl dump-flows xapi3 |grep icmp|grep 115.x.x.71"
> >
> > Flow appeared frequently lasting 1s and expired soon.Sometimes there
> were no flow for 10s
> >
> > Ping Outcome:
> > --- 115.x.x.17 ping statistics ---
> > 20 packets transmitted, 20 received, 0% packet loss, time 9012ms
> >
> >
> >
> >
> >
> >
> > On Thu, Jun 27, 2013 at 11:26 AM, Ben Pfaff <blp at nicira.com> wrote:
> > If you want clear answers, then you have to ask questions clearly.
> > As-is, I have to guess what you are really asking.
> >
> > On Thu, Jun 27, 2013 at 11:23:45AM +0530, ananthan wrote:
> > > thanks for your quick reply,but my ping didnt get dropped even though
> > > latency was varying
> > >
> > > On Thu, Jun 27, 2013 at 11:21 AM, Ben Pfaff <blp at nicira.com> wrote:
> > >
> > > > On Thu, Jun 27, 2013 at 11:20:20AM +0530, ananthan wrote:
> > > > >                   Could you please confirm this also.If there is no
> > > > > corresponding flow in data path how can packets reach with out any
> > > > slowness
> > > > > or being dropped.
> > > > > My understanding about data and control path is control path
> generate
> > > > flow
> > > > > for data path and if there is no matching flow in data path,
> control path
> > > > > has to regenerate one,in this case there is no flow for +_10
> seconds so
> > > > how
> > > > > can packets flow in this 10 seconds.
> > > >
> > > > If there is no flow for 10 seconds, then it indicates that there
> were no
> > > > packets during that time.
> > > >
> >
> > _______________________________________________
> > discuss mailing list
> > discuss at openvswitch.org
> > http://openvswitch.org/mailman/listinfo/discuss
>
> _______________________________________________
> discuss mailing list
> discuss at openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20130627/dcce17a3/attachment.html>