On Thu, Oct 25, 2012 at 2:48 PM, Peter Fales <peter.fa...@alcatel-lucent.com> wrote: > I'm trying to run cbench and against a local POX instance, but not having > much success. (Both POX and Oflops are fairly recent git clones). The > problem is that the POX application code never sees the Packet-in messages, > so it never sends back any flows to cbench. > > From a very quick attempt at debugging, it appears that POX does > not consider the switch to be fully "connected" until it has responded > to the first Barrier messages. And, cbench does not seem to handle > Barrier messages. There is a reference in the POX code to "an HP > switch that doesn't support barriers" but (reading between the lines) > those switches return a OFPET_BAD_REQUEST which POX treats as a > response to the Barrier. In the case of cbench, the Barrier is > dropped on the floor and no response of any kind is sent. > > Has anyone gotten cbench and POX to "play nice?"
Hi Peter, I don't know if anyone has run cbench on pox before but two points: 1) previous python controllers have gotten approx. 8k flow setups/second out of cbench 2) it would be pretty easy (~5 lines of code) to add a barrier reply handler to cbench if you were so inclined. Hope this helps, - Rob . _______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
