On 7/16/20 3:18 PM, [email protected] wrote: > Hi All, > > This is the OVN northd parallel processing, redone in small atomic > steps to make merging/rebasing easier. > > Patches 01-25 reorganize the processing in per-rule functions so > that it can be run in parallel. >
Hi Anton, This is not a complete review yet. I only looked at patches 01-25 until now. I have some comments on some of the patches but also a general comment: it might be better to use more explicit names that describe what part of the pipeline we're dealing with instead of "build_lrouter_flows_step_X_op/od". Same for logical switches. Thanks, Dumitru > Patches 26-28 move processing of lflows to a parallel model with everything > forced to run in parallel. > > Patch 29 makes processing for small deployments sequential switching > to parallel for larger ones. > > This patchset parallelises more aggressively than the first version. > The first version was parallelising only datapath and port processing > and synchronizing several times while doing it. This version also > tries to run in parallel load balancer and igmp group processing. It > also decreases the number of synchronization points where results > from different threads are merged into a common final result. > > It also fixes the sizing of the lflow hash which was broken and set to > a very low value at some point in the original patch series. > > Brgds, > > A. > _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
