Thanks for your prompt reply Mr. Lantz. I was busy with trying to update the experiment using --switch user. (See Benchmarking Mininet<http://vy.github.io/blog/post/2013/04/19/benchmarking-mininet/>post for the updated results.)
As you pointed out, user switch was performing much better than OVS in terms of memory consumption. But it does not cause much change on the timing results as anticipated. I interpret this finding as the bottleneck is caused by the issued ifconfig/route/ip calls. Thoughts? About the 1:1 switch ratio, since I am testing the maximum amount of switches that I attach to the system, attaching one host to each switch conforms with the benchmark objective. On Sat, Apr 20, 2013 at 1:28 AM, Bob Lantz <rla...@cs.stanford.edu> wrote: > Volkan, > > Have you tried --switch user also? OVS doesn't start up as quickly, > although it has better forwarding performance. > > Also the 1:1 switch:host ratio is probably a bit high for most real > networks at least. ;-) > > -Bob > > On Apr 19, 2013, at 4:10 AM, Volkan YAZICI <volkan.yaz...@gmail.com> > wrote: > > For those who are interested in, I put together a small mininet benchmark > that measures the memory usage and the start+shutdown time for a linear > topology composed of 1, 32, 64, 96, 128, ..., 1024 switch+host pairs. > Results are accessible from > http://vy.github.io/blog/post/2013/04/19/benchmarking-mininet/ address. > _______________________________________________ > openflow-discuss mailing list > openflow-discuss@lists.stanford.edu > https://mailman.stanford.edu/mailman/listinfo/openflow-discuss > > >
_______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
