Hi, On 2015年08月28日 08:44, Alan Deikman wrote: > >> On Aug 25, 2015, at 6:01 PM, Alan Deikman <[email protected] >> <mailto:[email protected]>> wrote: >> >> Can someone familiar with tester.py advise me as to a next step to take? > > I think I can explain the anomalies I reported in my prior e-mail. The > problem is that OVS when installed with the kernel datapath (DKMS) it > effective has a cache in the kernel for the flow table. The cached flow > entries have a hard timeout of 5 seconds. The problem is that tester.py > attempts to confirm its flow modifications with a barrier request, but OVS > sends a barrier request reply back when its non-cached table is stable but it > does not take note of the state of the cached entries in the kernel module. > > So as a result tester.py in some cases is throwing test packets at a switch > that is actually executing different flows than tester.py thinks it is. As > a work around I entered a time.sleep(10) statement right after the point at > which tester.py issues the clear command. With this modification running > with OVS 2.4.90 as both the test switch and the DUT I was able to obtain a > final score count of OK: 684 and ERROR: 307. This is 90 more tests that pass > without the mod and is a higher score than what is currently posted for OVS > on the Ryu certification page. > > At this point I am not sure if this should be considered an OVS bug or a > tester.py bug, but regardless anyone using tester.py against OVS as a test > switch should be aware of it.
If OVS work so, I think it is a bug of OVS and tester.py should not wait OVS. Because the OpenFlow Spec defines: --- Messages must not be reordered across a barrier message and the barrier message must be processed only when all prior messages have been processed. --- so OVS should not send a barrier reply if the flow table is not ready enough. This means that OVS can be missing the OpenFlow Spec. But for someone who want to pass the error case, adding time.sleep(10) maybe useful as a workaround. This is my personal view, though... Thanks, Iwase > > If someone would like to know more about my test results please feel free to > contact me. > > ---------------------------- > Alan Deikman > ZNYX Networks, Inc. > > > > > > ------------------------------------------------------------------------------ > > > > _______________________________________________ > Ryu-devel mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/ryu-devel > ------------------------------------------------------------------------------ _______________________________________________ Ryu-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ryu-devel
