The devices are setup using l2gw connections.
Now as we inject local ucast macs in one device using vtep-ctl command, they are appearing in remote ucast mac table of the other table.

Initial n (say 1000) macs are getting into second device at some rate.
After that the rate at which they are getting added is getting reduced.

The top cpu spending thread in the profiler analysis in mdsal notification thread.

For each mac that is being added, since there are listeners on Node level, the whole Node object is denormalized with all the macs and the Node update notification is generated.

This is unnecessary computation of Node object.

It is getting worse as the no of macs getting increased.

Now There are two ways this can be avoided.
1) Do not put a listener at Node level ( There are few listeners on Node level , all those have to listen on some other attribute like Global row/ Manager table to figure out a node is connected ).

2). Do not place ucast ( remote/local) under topology:hwvtep:1/node
Put them in a different container say topology:hwvtep:macs/node