On 2/13/25 17:03, David Marchand wrote: > On Thu, Feb 13, 2025 at 4:49 PM Mike Pattrick <[email protected]> wrote: >>> Thanks, Mike, for the additional testing. So, it seems there is a trade-off >>> between some performance decrease in basic V2V cases and a sizeable increase >>> in cases with the actual recirculation involved. >>> >>> So, I'd say it seems fine to me to make that trade as more setups have more >>> complex pipelines these days. >>> >>> What do you think? >>> >>> For the backports though, I don't think this should go below 3.5, as sudden >>> changes in performance may not be desired on stable branches. Also, the >>> feature is still experimental and it's more of a known behavior rather than >>> a bug that we drop TSO packets on recirculation. >> >> I think that's reasonable! > > Previous versions of OVS were lacking other bits for TSO, like the > software fallback. > So backporting only to 3.5 looks fine for me. > > Btw, when this patch gets applied, could you add a tag in the commitlog? > I have an internal bug tracking this issue. > Reported-at: https://issues.redhat.com/browse/FDP-1144 > > I can send a new revision if you prefer.
Thanks, David and Mike! I added the tag and applied the change to main and 3.5. Best regards, Ilya Maximets. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
