Asynchronous cross-site replication relies on the ordering of the write operation to ensure data consistency. This ordering is provided by JGroups; it only delivers the next request after the first one is completed.

This order contains updates for multiple caches and keys. What is proposed here is to handle different keys/cache updates in parallel.

It should improve the response time, however, there is no free lunch. The following is expected:

• An increase of CPU utilization: it may spawn new threads to handle different keys update in parallel
• An increase in memory: it chains the updates (queue)