Hi IIlya, > -----Original Message----- > From: Ilya Maximets <[email protected]> > Sent: Saturday, March 12, 2022 3:26 AM > To: Amber, Kumar <[email protected]>; [email protected] > Cc: [email protected]; Phelan, Michael <[email protected]> > Subject: Re: [ovs-dev] [PATCH v1] system-dpdk.at: Add ignore warning for > context switches in CI. > > On 3/2/22 10:42, Amber, Kumar wrote: > > Hi llya, > > s/LL/IL/ > > > > >> -----Original Message----- > >> From: Ilya Maximets <[email protected]> > >> Sent: Tuesday, March 1, 2022 1:50 AM > >> To: Amber, Kumar <[email protected]>; [email protected] > >> Cc: [email protected]; Phelan, Michael <[email protected]> > >> Subject: Re: [ovs-dev] [PATCH v1] system-dpdk.at: Add ignore warning > >> for context switches in CI. > >> > >> On 2/28/22 19:23, Amber, Kumar wrote: > >>> Hi IIya, > >>> > >>> Thanks for the comments > >>> > >>>> > >>>> > >>>> Hi. These warnings are more about long poll intervals and that > >>>> should not happen during tests. Which test is causing this problem? > >>>> > >>>> Long poll intervals usually indicate a problem with the test itself. > >>>> Usual suspect is poor pinning of PMD threads. > >>>> > >>>> Best regards, Ilya Maximets. > >>> > >>> Fuzzy testing for MFEX , I agree but the cores of the test were > >>> chosen such > >> that the cores are always available on any available Setup. > >>> Thus, this is a side-effect, but it guarantees that the test won't > >>> fail on most of the setups. Suppressing warnings is A good tradeoff. > >> > >> Why can't we use dummy numa for this test? This will avoid actual > >> pinning of the threads keeping them available for kernel scheduling. > >> We're not testing performance here after all. > >> > >> Best regards, Ilya Maximets. > > > > Based on the discussion and feedback, we have tried to remove the complexity > of the DPIF itself from testing MFEX. > > To Make it simpler to understand and introduce testing more complex traffic > type and improvements to the testing of mfex. > > > > Please have a look at the patch sets and feedback is always great to improve > testing. > > http://patchwork.ozlabs.org/project/openvswitch/list/?series=287947 > > Sure, I'll take a look. > But on a quick glance I see that these system tests are not removed by the > mentioned patch set. So, I'm not sure how it is related to the current > discussion. > > Best regards, Ilya Maiximets.
Yes, I totally agree with the Performance aspect that we ae not testing it and that’s why we are also trying to find ways to improve testing. Until then it's important for us to enable this test in AVX512 CI and that why it’s a quick fix to enable the test. Thanks a lot for the Feedback will consider on the other patch and will produce something more concrete. Till then if this patch can be merged, the test will be running in CI once we have something better which is accepted to community will remove the test altogether. Regards Amber _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
