Improved Ehcache performance in large clusters

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 5.1 EE SP4 (5.1.7), 5.2 EE SP3 (5.2.7)
• Fix Version/s: 5.1 EE SP5 (5.1.8), 5.2 EE SP4 (5.2.8)
• Component/s: Fault Tolerance > Clustering Framework, Performance
• Labels:
  • 5.2_release_notes
• Environment: All

Ehcache's build-in replication mechanism creates a thread for each cache. Liferay portal requires over 200 caches which means each cluster node has 200+ Ehcache replication threads. In addition, Ehcache uses RMI to communicate with other nodes. These two issues makes Ehcache inefficient in a large cluster environment.

This improvements switches Ehcache's replication mechanism with a new replication mechanism built on ClusterLink. The number of threads need on each cluster node is reduces to 10-20 threads and communication is now node using UDP(JGroups).

To enable the new replication mechanism:

1) Because the replication mechanism uses cluster link, set the following in portal(-ext).properties;
cluster.link.enabled=true

2) Add the following to the "spring.configs" property in portal(-ext).properties
META-INF/cache-cluster-spring.xml

3) Modify Ehcache's cluster configuration files(hibernate-clustered.xml and liferay-multi-vm-clustered.xml).

A) Remove the cacheManagerPeerProviderFactory and
cacheManagerPeerListenerFactory tags. These are not needed any more.

B) Inside each defaultCache and cache entry, replace the following
<cacheEventListenerFactory ... />
<bootstrapCacheLoaderFactory class="net.sf.ehcache.distribution.RMIBootstrapCacheLoaderFactory" />

with a single entry

<cacheEventListenerFactory class="com.liferay.portal.cache.cluster.EhcachePortalCacheClusterReplicatorFactory" />