[ovs-discuss] OVS switching capacity vs. cpu utilization

Faizul Bari faizulbari at gmail.com
Thu Sep 5 03:52:34 UTC 2013


Thanks a lot for the info. I have looked at the paper you mentioned.

I want to plot "Transfer Rate (mbps)" on the x-axis and on the Y-axis I
want to measure the "CPU utilization". I am using UDP flows and measuring
the CPU consumption of the process ovs-vswitchd. Is this the kernel module
you mentioned in your email?

--
Faiz


On Wed, Sep 4, 2013 at 6:30 PM, Henrique Rodrigues <henriquesilvar at gmail.com
> wrote:

> One of the first papers about ovs have something similar to what you want
> to do: http://openvswitch.org/papers/dccaves2010.pdf They got ~7G at that
> time, depending on the transfer (pkt) size.
>
> I believe you should also try UDP and non ip traffic, to isolate the
> tcp/ip stack code. OVS also has a kernel module, so using something else to
> check how much cpu the kernel module consumes would be interesting as well.
>
> Another thing you should try is to search for "openvswitch performance" on
> google scholar. This would give you some other interesting reading sources.
>
>
> On Thu, Aug 29, 2013 at 3:01 PM, Faizul Bari <faizulbari at gmail.com> wrote:
>
>> Hello,
>>
>> I am trying to measure the switching capacity of OVS. I am trying to plot
>> cpu utilization against amount of traffic (mbps) switched by OVS.
>>
>> I have deployed OVS on a workstation and connected to two other other
>> workstations running iperf server and client. Now which process on the OVS
>> workstation should I monitor? Is it ovs-vswitchd?
>>
>> Is there any existing such measurement?
>>
>> --
>> Faizul Bari
>>
>> _______________________________________________
>> discuss mailing list
>> discuss at openvswitch.org
>> http://openvswitch.org/mailman/listinfo/discuss
>>
>>
>
>
> --
> Henrique Rodrigues
> http://www.dcc.ufmg.br/~hsr
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20130904/bdbae75c/attachment.html>


More information about the discuss mailing list