On Thu, Jun 24, 2021 at 10:33 AM Dumitru Ceara <[email protected]> wrote: > > On 6/24/21 7:20 PM, Han Zhou wrote: > > For the reason mentioned above, we can't make this change. In fact, I > > wouldn't worry much about lflow_ref_lookup()'s cost. It is O(1) operation. > > If it really turns out to be a bottleneck, we could optimize the > > function/data-structure, without worrying about the logic. > > The real performance impact part is probably not being able to cache the > > "match" for lflows that have logical port references, but I will work on > > some other solutions to optimize that. > > OTOH, on real deployments the lflow cache limits should be enforced by > the CMS. Therefore I would expect some of these flows to not make it in > the cache anyway (even without your change). I don't have data to back > this up but I'm guessing the impact of the change in this patch will be > minimal. > > Regards, > Dumitru >
Thanks Dumitru. Numan, I sent v4 that adds more coverage in the test case. Please take a look: https://patchwork.ozlabs.org/project/ovn/patch/[email protected]/ Thanks, Han _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
