ARP requests are both replicated to all ports by means of a multicast group, and cloned to ONOS
ARP replies, unicast forwarding + clone to ONOS
IPv4/ICMP etc, unicast forwarding between the two hosts
What's the relationship with the L3 table? Probably it's based on the l3 classifier table, if hit, then process via L3 pipeline, otherwise use L2 pipeline.
Learning happens in ONOS
What's the relationship with the L3 table? Probably it's based on the l3 classifier table, if hit, then process via L3 pipeline, otherwise use L2 pipeline.