Hi santosh,

1.      In HA environment i.e. VTEP redundancy  ,  ODL operating in A/A mode ( referring to your document ) ,  ODL takes care of not updating the remote-ucast-mac entries to the peer H/W VTEP with local_ucast_mac entries learnt from its peer.  With NSX we are observing this behavior .

Daya->yes, correct, ODL keeps track of which 2 hwvteps are peers, and does not try any sync between them. it assumes the peers will share tunnel_ip’s, and have MLAG type connectivity to the compute nodes, so that the mac’s will be synced by other channels.

2.      Can't we use vtep ips to form the HA node in case of A/A mode . i.e.  H/W vteps publishing same vtep_ips are part of the same cluster. ?  Was this approach  thought of  ?

Daya->yes, this could be done, but we wanted to avoid any implicit assumptions, and also guard against mis-configs, or have incorrect behavior during transitions. For example, if a node is removed, and replace by another one with the same vtep ip, this should not be treated as an HA configuration.

3.  Remote macs entries would be flushed only in response to the delete notification for mac deletion from local_ucast_table   and not in response connectivity loss between Hardware VTEP and the ODL controller.  Is it correct assumption  ?

Daya->am tempted to say correct, on paper, but need to test out what the re-conciliation logic is doing currently, there were some fixes that went in recently in this code. Will check and update.

Thanks,

daya