Hello Han, Dumitru, Numan, and others,
As per OVN architecture documentation, `join` logical switch is exclusively used to connect the gateway routers to distributed routers. So, all the ports connected to the join switch are ‘router’ logical_switch_ports. Now, 1. Say a packet arrives on the gateway router and is meant for a VM behind a logical switch. We lookup the routing table, determine the nexthop, find the destination MAC address for the nexthop IP from lr_in_arp_resolve table and send the packet out towards the distributed router. 1. On the other direction, the packet from a VM towards 8.8.8.8 arrives at the distributed router. We lookup the routing table, determine the nexhop, find the destination MAC address for the nexthop IP lfrom lr_in_arp_resolve table and send the packet out towards the gateway router. >From (a) and (b) above, at least on the Join switch, we should never see any ‘ARP requests at all, right? If this is the case, then the whole ls_in_arp_rsp table for the join switch can be just table=13(ls_in_arp_rsp ), priority=0 , match=(1), action=(next;) Any thoughts? In a 1000 Node K8s cluster, there are around 12K Logical Flows for `join` switch. Out of which, 6K Logical flows are from table=13 and none of those flows are being used. I am trying to see if we can optimize this path. Regards, ~Girish
_______________________________________________ discuss mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
