Hey Nikhil,

A few things to consider.

- Nox is *not* a very efficient controller. The event model using argument binding and function pointers adds a lot of overhead, and significant time is wasted checking fd's which don't have any events pending. There are other scalability limitations as the number of active OpenFlow connections increases over a few hundred. If you want to test against a more efficient controller, the trivial C controller shipped with the reference solution and Open vSwitch is probably your best bet.

- Both event processing latency, and throughput are useful metrics for an OpenFlow controller. Latency provides and indication of how fast a controller can respond to a single event. And throughput is useful to understand how the controller handles load. Amin Tootoonchian has done some pretty interesting latency work with Nox in which he was able to get something like 17us end-to-end latency using a tickless kernel with high-fidelity timers. Throughput with Nox is definitely CPU limited and can only be test with multiple active OpenFlow connections.

- Single instance throughput is probably not a very interesting metric for most real deployment environments. Latency, and the ability for the controller to scale-out are more germane measurements. Testing the latter is difficult because it converges on the consistency overhead of the control logic implemented using the controller, and not the controller platform itself.

.martin

Hi,

I'm looking for a way to do an apples-to-apples comparison of the performance of various controllers out there. From what I have heard NOX is a very efficient controller with some impressive performance numbers.

Are there standard benchmark tools that are used to evaluate the performance of NOX? If so, are they publicly available?

Thanks,
Nikhil


_______________________________________________
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

Reply via email to