Hi Zheng, You can delete flows matching only at the output port. You could use this to make sure that there aren't any "noop" flows after a link failure.
Thanks, Yiannis On Sun, Jan 3, 2010 at 10:47 PM, kk yap <[email protected]> wrote: > Hi Zheng, > > Comments inline. > > > I was wondering what would be the behavior of the NOX controller(using > > routing app package) and of an Openflow switch upon link failures. > > > > To be more specific, assume that there are already some flow entries(FEs) > in > > switch A which will output packets to port X (X is connected to port Y on > > switch B). When this link fails, I guess that both A and B will send > > ofp_port_status messages to the controller. My question then is, what > will > > happen to these flow entries(FEs)? Will they be removed by the switch > itself > > immediately, or they have to wait for the hard timeout? During that > period, > > will all the packets matched by those entries be going to the dead link? > > AFAIK, the entries will not timeout. If the flows are TCP or some > end-to-end controlled protocol that timeouts, then it will stop > sending packets. Then the entries will timeout due to idle (not hard > timeout). I believe in most hardware you have lost those packets. > > However, it should not be hard to poison these entries when NOX sees a > link timeout. I am assuming that is what you are asking. Your > experience with discovery might vary though. > > > To my knowledge, the NOX routing app package does not remember the flow > > entries it create(correct me if I am wrong). Then if it is the hard > timeout > > case, it seems that it will take some time for those entries to get > > corrected by the controller. I really appreciate it if someone can point > me > > to a place where I can figure out this whole process. > > Yes, routing do not remember the flow entries. If you want to > understand the process, read the source code discovery, topology, > authenticator and routing. The last being the most important of > course. Unfortunately documentation in NOX lags significantly. NOX > 0.6 has adopted doxygen, and hopefully documentation will catch up. > > Hope the comments are helpful. > > Regards > KK > > _______________________________________________ > nox-dev mailing list > [email protected] > http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org > -- Yiannis .
_______________________________________________ nox-dev mailing list [email protected] http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
