Hi Rob, Ben and Wanderson, Thank you all for your reply.
OVS does have a good performance, and I am not trying to avoid OVS. I want to use some extenal hardware to improve PC-based OpenFlow Switching performance, which is quite similar to this article "Using Hardware Classification to Improve PC-based OpenFlow Switching". They used a NIC hardware classification to offload the CPU from the lookup processing with OpenFlow kernel module version 0.8.9-rev4. I asked the author, and we believe that we need to use OpenFlow implementation that support at least version 1.0 for further experiment from now on. Currently, the kernel module implementation that support OpenFlow 1.0 is only available through ovs. So, I want to know whether ovs can be treat as a baseline, since we only take into account the OpenFlow switching, and ovs offers more features beyond this. Moreover, when will ovs fully support OpenFlow 1.1? is there a plan? Hoping i've make it clear, Thanks a lot. Best Regards, HYN Hi, Huyn, i would suggest you to read this article "OpenFlow Switching: Data Plane Performance". Is easy to find if you google it. They use a PC-based OpenFlow Switching in order to test the performance of the Data Plane. Maybe it is what you are looking for. The use of OVS is not clear in the article, but the results may provide you an idea about the performance. Wanderson 2011/10/17 Ben Pfaff <[email protected]> On Mon, Oct 17, 2011 at 11:07:31AM -0700, Rob Sherwood wrote: > 2011/10/17 huyn <[email protected]>: > > Is there other kernel module OpenFlow implementations besides OVS? > > > > If not, can we use OVS to test the PC-based OpenFlow Switching performance? > > Or should we implement a pure kernel module OpenFlow Switch ourselves?? > > The stanford reference switch has a kernel implementation, but it was > deprecated with the 1.0 version. There was an attempt to update the > kernel version to 1.0 and I'm not sure how far it got, but IMHO, if > you're interested in performance, you should just use OVS. I agree, but my reading of huyn's message is that he wants to avoid OVS. huyn, is that true, and if so, why is it the case? Thanks, Ben (one of the authors of OVS). _______________________________________________ openflow-discuss mailing list [email protected] https://mailman.stanford.edu/mailman/listinfo/openflow-discuss -- Wanderson Paim de Jesus Msc Student in Computer Science 51 8519 9136 www.inf.ufrgs.br/~wpjesus
<<inf(10-18-12-35-27).jpg>>
_______________________________________________ openflow-discuss mailing list [email protected] https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
