One last question, is there any way to measure how many flows are installed in a flow table without using any script to count the output of the ovs-ofctl dump-flows command?becase while it's counting it adds some delay and the results are not exact.
Thank you. 2011/5/25 Jose Luis Franco Arza <[email protected]> > Yes, now the performance it's much better, thank you!at least now it > installs all the flows thuogh when they are installed then it starts to lose > some flows but I guess that this is because of the idel timeout. I will try > it increasing it to check if is there still loss of flows. > Thanks again, > > Greetings, > > Jose > > > 2011/5/25 Jose Luis Franco Arza <[email protected]> > >> Thank you very much Ben, I will try it now and I will let you know >> >> >> 2011/5/25 Ben Pfaff <[email protected]> >> >>> On Wed, May 25, 2011 at 07:19:53PM +0200, Jose Luis Franco Arza wrote: >>> > > >>> > > That looks like OFPET_BAD_REQUEST, OFPBRC_BUFFER_UNKNOWN "Specified >>> > > buffer does not exist". I imagine that on the switch side you got a >>> > > bunch of "cookie mismatch" log messages. >>> > > >>> > >>> > Exactly, in the openvswitch I receive a lot of cookie mismatch messages >>> and >>> > also some other log message that I don't understand at all (I read that >>> they >>> > were used for debuggin openvswitch so I just ignore them) >>> >>> You should be able to find a line like this in ofproto/connmgr.c in >>> Open vSwitch: >>> ofconn->pktbuf = pktbuf_create(); >>> >>> If you delete that line, do you see better performance? If so, then >>> perhaps it's time to work on making OVS suppress packet buffering in >>> overload conditions, or maybe to just drop support entirely. >>> >>> We don't see this problem at Nicira because our controllers set up >>> flows proactively, not reactively. >>> >> >> >
_______________________________________________ openflow-discuss mailing list [email protected] https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
