Another thought, often we've been limited by the throughput of the
switch. We've had to run up to 4 switches to max out the setup capacity
of the controller. Have you verified whether you are CPU bound on the
controller?
C++ only is required for high throughput. Looking through the last
performance numbers I have (about 4 months old) it looks like
throughput maxed out closer to 40k/s. I not sure what the discrepancy
is, we've modified the core events loop a number of times.
With Python we never reached more than 13-16k flows/s.
btw, if you use schedtool to bind the process to a single core on a
multicore machine, you should see an increase in performance in general.
.martin
Hi, Martin
I saw your email in this nox-dev list, saying that you can get 60K
flow setup per second.
http://www.mail-archive.com/[email protected]/msg00174.html
I wonder what your experiment setup is. How do you send the flows?
Which hardware do you run nox?
I try to run nox with routing and pyswitch module. I can only see
6K-7K flows/sec with nox in an Intel xeon machine. My sender and
openflow switch are in different machines. I wonder if it is my
experiment setup problem.
Thanks,
-- Minlan
_______________________________________________
nox-dev mailing list
[email protected]
http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
_______________________________________________
nox-dev mailing list
[email protected]
http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
