Problem:
Both, cache replication and cluster nodes discovery in EhCache are sequential and synchronous.
Time required to replicate values is proportional to number of nodes in cluster.

com.atlassian.cache.ehcache.replication.rmi.RMISynchronousCacheReplicator is a wrapper for net.sf.ehcache.distribution.RMISynchronousCacheReplicator

We could use RMIAsynchronousCacheReplicator but it introduces numerous problems with backpressure and keeping soft references to objects waiting for replication.

Solution:
Implement semi-async replication where:

• fire async replication tasks for each node
• block until all tasks are done

Implement parallel cluster nodes discovery

It would significantly reduce cache replication time for large clusters