On 04/05/2021 17:12, Yogananth Subramanian wrote: > Hello Everyone Hi Yogananth,
> I was able to verify the patch, below are the observed behavior of OVS > 2.15 release and OVS from git with the patch. > OVS 2.15 release : > > PMD auto balance aborted: > > 2021-04-27T08:54:54.770Z|00406|dpif_netdev|DBG|PMD auto lb dry run. > Current: Core 3, usage 51 > > 2021-04-27T08:54:54.770Z|00407|dpif_netdev|DBG|PMD auto lb dry run. > Current: Core 5, usage 51 > > 2021-04-27T08:54:54.770Z|00408|dpif_netdev|DBG|PMD auto lb dry run. > Current: Core 27, usage 96 > > 2021-04-27T08:54:54.770Z|00409|dpif_netdev|DBG|PMD auto lb dry run. > Current: Core 29, usage 93 > > 2021-04-27T08:54:54.770Z|00410|dpif_netdev|DBG|PMD auto lb dry run. > Aborting due to cross-numa polling. ^ right, this is where without the patch any cross-numa polling prevents the alb feature from working. > OVS from git with PDM cross numa rebalance commit: > > PMD auto balance triggered: > > 2021-04-27T11:32:17.985Z|00805|dpif_netdev|INFO|PMD auto lb dry run. > requesting datapath reconfigure. > > 2021-04-27T11:32:17.985Z|00806|dpif_netdev|WARN|There's no available > (non-isolated) pmd thread on numa node 0. Queue 0 on port 'dpdk-p1' will be > assigned to the pmd on core 3 (numa node 1). Expect reduced performance. > > 2021-04-27T11:32:17.985Z|00807|dpif_netdev|WARN|There's no available > (non-isolated) pmd thread on numa node 0. Queue 0 on port 'dpdk-p0' will be > assigned to the pmd on core 29 (numa node 1). Expect reduced performance. > > 2021-04-27T11:32:17.985Z|00808|dpif_netdev|WARN|There's no available > (non-isolated) pmd thread on numa node 0. Queue 2 on port 'dpdk-p0' will be > assigned to the pmd on core 27 (numa node 1). Expect reduced performance. > > 2021-04-27T11:32:17.985Z|00809|dpif_netdev|WARN|There's no available > (non-isolated) pmd thread on numa node 0. Queue 1 on port 'dpdk-p0' will be > assigned to the pmd on core 5 (numa node 1). Expect reduced performance. > > 2021-04-27T11:32:17.985Z|00810|dpif_netdev|WARN|There's no available > (non-isolated) pmd thread on numa node 0. Queue 2 on port 'dpdk-p1' will be > assigned to the pmd on core 5 (numa node 1). Expect reduced performance. > > 2021-04-27T11:32:17.985Z|00811|dpif_netdev|WARN|There's no available > (non-isolated) pmd thread on numa node 0. Queue 1 on port 'dpdk-p1' will be > assigned to the pmd on core 27 (numa node 1). Expect reduced performance. > ^ yes, this indicates that the alb has triggered a rebalance while there is cross-numa polling limited to pmds on one numa node, which is the intended behaviour. btw, just to note for future mails that plaintext is preferred on the mailing list. Thanks for testing, Kevin. > > Before rebalance > > After rebalance > > pmd thread numa_id 1 core_id 3: > > isolated : false > > port: dpdk-p1 queue-id: 0 (enabled) pmd usage: 46 % > > pmd thread numa_id 1 core_id 5: > > isolated : false > > port: dpdk-p0 queue-id: 0 (enabled) pmd usage: 32 % > > port: dpdk-p0 queue-id: 1 (enabled) pmd usage: 26 % > > pmd thread numa_id 1 core_id 27: > > isolated : false > > port: dpdk-p0 queue-id: 2 (enabled) pmd usage: 35 % > > port: dpdk-p1 queue-id: 2 (enabled) pmd usage: 18 % > > pmd thread numa_id 1 core_id 29: > > isolated : false > > port: dpdk-p1 queue-id: 1 (enabled) pmd usage: 13 % > > pmd thread numa_id 1 core_id 3: > > isolated : false > > port: dpdk-p1 queue-id: 0 (enabled) pmd usage: 76 % > > pmd thread numa_id 1 core_id 5: > > isolated : false > > port: dpdk-p0 queue-id: 1 (enabled) pmd usage: 47 % > > port: dpdk-p1 queue-id: 2 (enabled) pmd usage: 23 % > > pmd thread numa_id 1 core_id 27: > > isolated : false > > port: dpdk-p0 queue-id: 0 (enabled) pmd usage: 67 % > > port: dpdk-p1 queue-id: 1 (enabled) pmd usage: 19 % > > pmd thread numa_id 1 core_id 29: > > isolated : false > > port: dpdk-p0 queue-id: 2 (enabled) pmd usage: 54 % > > > > On Mon, Mar 22, 2021 at 5:31 PM Ilya Maximets <i.maxim...@ovn.org> wrote: > >> On 3/18/21 12:34 PM, Kevin Traynor wrote: >>> Previously auto load balance did not trigger a reassignment when >>> there was any cross-numa polling as an rxq could be polled from a >>> different numa after reassign and it could impact estimates. >>> >>> 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. Allow PMD auto load balance to trigger a reassignment in >>> this case. >>> >>> Acked-by: Eelco Chaudron <echau...@redhat.com> >>> Acked-by: David Marchand <david.march...@redhat.com> >>> Tested-by: Sunil Pai G <sunil.pa...@intel.com> >>> Acked-by: Flavio Leitner <f...@sysclose.org> >>> Signed-off-by: Kevin Traynor <ktray...@redhat.com> >>> --- >>> https://github.com/kevintraynor/ovs/actions/runs/664214192 >>> v3: >>> - Updated docs and logs as per Ilya suggestion >>> v2: >>> - Same logic as v1, combined two "ifs" as per David suggestion >>> - Updated comments/logs >>> - Updated the doc note that said it will not work for cross NUMA to >>> include new condition >>> - Kept Eelco's Ack, as no logic changed >>> --- >>> Documentation/topics/dpdk/pmd.rst | 4 +++- >>> lib/dpif-netdev.c | 19 ++++++++++++++++--- >>> 2 files changed, 19 insertions(+), 4 deletions(-) >> >> Thanks! I added a NEWS entry and applied to master. >> >> Best regards, Ilya Maximets. >> _______________________________________________ >> dev mailing list >> d...@openvswitch.org >> https://mail.openvswitch.org/mailman/listinfo/ovs-dev >> >> > _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
