Li: New Connections are slower in a cluster as txns are not batched across devices.

Currently when a device connects, a txnChain is created for that device.

Each device connection in a minimum creates 2 txns , 1 for inventory and other for topology.

Now in a network with few switches, this would fare fine, but when we have a network with 1000 switches connected across all the nodes of a 5 node cluster,
the number of txns increase.
This causes heavy replication traffic across east-west and turnaround time of a new connection goes down.

The He inventory-manager (as i recollect) used to batch the txns accross devices. We would need a similar model here to scale.