Appendix F. Performance Recommendations

F.1. Concurrent Startup for Large Clusters

When starting a large number of instances, each managing a large number of caches, in parallel this may take a while as rebalancing attempts to distribute the data evenly as each node joins the cluster. To limit the number of rebalancing attempts made during the initial startup of the cluster disable rebalancing temporarily by following the below steps:

  1. Start the first node in the cluster.
  2. Set JMX attribute jboss.infinispan/CacheManager/"clustered"/LocalTopologyManager/rebalancingEnabled to false, as seen in LocalTopologyManager.
  3. Start the remaining nodes in the cluster.
  4. Re-enable the JMX attribute jboss.infinispan/CacheManager/"clustered"/LocalTopologyManager/rebalancingEnabled by setting this value back to true, as seen in LocalTopologyManager.