Every release has improved performance, so I'd recommend the latest (1.10.x).  
If you want the long-term support version, then 1.9.x.  hping is a particularly 
difficult use-case, since it sets up lots of different short-lived flows 
instead of sending lots of packets through just a few flows.  The hping 
use-case should be significantly better once kernel wildcarding is in.  Patches 
for that should go out in the next couple of weeks, which you may want to test.

--Justin


On May 29, 2013, at 3:36 AM, kevin parker <[email protected]> wrote:

> Hi Justin,
>              Which version do u recommend,which is capable enough to use low 
> cpu:
> 
> I tested on  kvm host1 running OVS 1.6.90 and created vm-1  and on another 
> kvm host2 created  2 vms (vm1 and vm2) then from vm2 and vm3 using hping3 i 
> flooded vm1.
> 
> initially flow was 500 later came down to <5 
> 
> sudo ovs-dpctl show
> 
> system@kvmbr0:
>       lookups: hit:29144 missed:71485237 lost:8866
>       flows: 3
>       port 0: xenbr0 (internal)
>       port 1: eth1
> 
> but CPU usage of ovs-vswitchd was ~100% when i increased  
> flow-eviction-threshold to 2000
> 
>  CPU usage did not decrease but total RAM usage increased by 10% .
> On other vms running on host1 there was more than  25% packet loss for ping 
> request and  broken pipe for ssh session.
> 
> 
> Regards,
> Kevin
>  
> 
> 
> On Tue, May 28, 2013 at 10:46 PM, Jing Ai <[email protected]> wrote:
> 
> > Date: Mon, 27 May 2013 09:10:20 -0700
> > From: Justin Pettit <[email protected]>
> > Subject: Re: [ovs-discuss] HIgh CPU usage for ovs-vswitchd with flows
> > 3069        and lost: 267425491
> > To: kevin parker <[email protected]>
> > Cc: [email protected]
> > Message-ID: <[email protected]>
> > Content-Type: text/plain; charset=iso-8859-1
> > 
> > We've made a lot of improvements in flow set up rate since version 1.4, so 
> > upgrading to a more current version (we're on 1.10 now) will likely help. 
> > We're currently working on multithreading the OVS userspace and adding 
> > support for wildcarded flows in the kernel, which should substantially 
> > improve flow set up.
> > 
> > --Justin
> > 
> 
> Good to know there are ongoing efforts to improve flow setup. What are the 
> timelines for those features? Any patch sent out for review? Thanks!
> 
> Best,
> Jing
> 
> 
> > 
> > On May 27, 2013, at 12:59 AM, kevin parker <[email protected]> 
> > wrote:
> > 
> > > Hi,
> > > 
> > > Running OVS 1.4 on xenserver 6.0.2 , but its taking very high cpu some 
> > > times ~100%.
> > > 
> > > ovs-dpctl show
> > > 
> > > system@xenbr5:
> > > lookups: hit:2560723 missed:3742809 lost:0
> > > flows: 5
> > > port 0: xenbr5 (internal)
> > > port 1: eth5
> > > system@xapi2:
> > > lookups: hit:1660559495 missed:1241428 lost:0
> > > flows: 11
> > > port 0: xapi2 (internal)
> > > port 1: eth7
> > > port 2: eth6
> > > system@xenbr4:
> > > lookups: hit:2539909 missed:3729876 lost:0
> > > flows: 5
> > > port 0: xenbr4 (internal)
> > > port 1: eth4
> > > system@xapi3:
> > > lookups: hit:20443295213 missed:26596588140 lost:267425491
> > > flows: 3069
> > > port 0: xapi3 (internal)
> > > port 1: eth1
> > > port 2: eth0
> > > port 4: xapi4 (internal)
> > > port 15: vif12.0
> > > port 18: vif14.0
> > > system@xenbr2:
> > > lookups: hit:1634980795 missed:166104910 lost:0
> > > flows: 127
> > > port 0: xenbr2 (internal)
> > > port 1: eth2
> > > system@xenbr3:
> > > lookups: hit:2450949145 missed:81360495 lost:0
> > > flows: 118
> > > port 0: xenbr3 (internal)
> > > port 1: eth3
> > > port 2: xapi6 (internal)
> > > port 6: vif12.1
> > > port 8: vif14.1 
> > > 
> > > Network usage:
> > > 
> > > dstat -n
> > > 
> > > -net/total-
> > > recv send
> > > 6475k 5736k
> > > 6575k 5646k
> > > 6767k 6347k
> > > 
> > > Can some one please tell me how this can be fixed.
> > > 
> > > Regards,
> > > Kevin
> 
> 

_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to