HA replication for persistent operatopnal data


We are planning to keep some operational data, more on current status of the system in many areas, that can be voluminous.

For example, L2 learning – this table can be huge.

The number of sources for this information being more than one (in this example case, anyone that processes ARPs, sniffs incoming or outgoing packet), the choices are
one: we write a daemon that collects & caches info from multiple sources and distribute
two: Use confd as that central person with persistent operational data in CDB and use Confd’s notification mechanisms.

If we pick two (use confd operational data)

  1. Any size limitation(s), we should be aware of ?
  2. In a clustered system, where we keep a master confd and a slave-confd on every other blade, what would be the replication overhead, both in synchronous & asynchronous modes ?
  3. What are the performance bottlenecks that we should take into consideration ?

Thank you in advance for your time & help!