On 9/27/23 17:33, Mark Michelson wrote: > On 9/27/23 07:15, Dumitru Ceara wrote: >> On 9/26/23 19:26, Mark Michelson wrote: >>> >>> 2) I don't think a controller action is necessary. The controller action >>> sets some registers and then resubmits to table 40 or 41. I think this >>> could be done directly in OpenFlow instead. >> >> Hi Mark, >> >> I didn't review the patch at all yet so I'll just comment on this >> specific point: >> >> The controller action is actually necessary in order to "break" the >> openflow rule chain that gets executed for packets that need to be >> flooded on these large multicast groups. >> >> Otherwise ovs-vswitchd will "combine" those OpenFlow rules into a single >> (or a handful of) datapath flow(s) with a huge action set that >> sequentially outputs to all ports in the multicast group. > > Is the problem that the megaflow's giant action set still overflows the > netlink maximum action size? >
I think, in theory, the resulting megaflow is exactly the same as it would be without this patch applied. _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
