+++ This bug was initially created as a clone of Bug #1938162 +++ PMD auto load balance does not trigger a reassignment when there is any cross-numa polling as an rxq could be polled from a different numa after reassign so the estimates it makes for improvement after reassignment may be incorrect. In the case where there is only one numa with pmds available, the same numa will always poll before and after reassignment, so estimates are valid and auto load balance could be allowed. There are customer cases where they are configuring like this and want to use PMD auto load balance. Allow PMD auto load balance to trigger a reassignment in this case. --- Additional comment from Kevin Traynor on 2021-03-16 10:59:27 CET --- v2 patch: https://mail.openvswitch.org/pipermail/ovs-dev/2021-March/381237.html --- Additional comment from Kevin Traynor on 2021-03-16 16:43:02 CET --- Some notes for QE validation. It can be tested with physical NICs or VMs or a combination. The key things needed for a test are: - there is at least one interface on a numa node that has no available pmd cores - there is only one numa with available pmd cores plus the normal pmd auto load balance conditions - some core is heavily loaded with multiple rxqs while another one is not - auto load balance is enabled With this patch, and the conditions above, auto load balance should occur, where previously it would not be permitted. --- Additional comment from OvS team on 2021-03-22 21:19:47 CET --- * Mon Mar 22 2021 Kevin Traynor <ktraynor> - 2.13.0-98 - dpif-netdev: Allow PMD auto load balance with cross-numa. (#1938162) [edeaca020b9d40fed8a0a474f35eeec5ed4e338f]
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory (openvswitch2.13 bug fix and enhancement update), and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2021:1055