WildFly-8.2 clustering - Best Method?

[valsaraj007]

Hi,

I found in a blog that there are two ways to build a cluster that is both: highly availably and scalable. The first way will use a concept we call sub-clusters and the second way will use a feature of the Infinispan-cache called distribution.

Which one is the best method?

Thanks!

[pferraro]

I assume the blog to which you are referring is Scalable HA Clustering with JBoss AS 7 / EAP 6 | akquinet AG – Blog ?

There is no single answer to your question.  Each strategy has its advantages and disadvantages.

The multi-cluster w/replication approach (sub-cluster is a bad name, since a "sub-cluster" is a distinct cluster) is slightly more complex to scale since it requires a distinct server-group (with unique load-balancing group) per cluster, however the cost of adding/removing a member to/from a cluster is fairly small, since the transfer of state to the new member does not affect other clusters.  It offer potentially higher availability (provided no cluster loses all members), but is less able to spread load evenly, since a given piece of state is only available within 1 cluster.

Contrarily, the single-cluster w/distribution approach is simpler to scale, since you only need a single server-group/load-balancing-group, however adding/removing a member to/from the cluster requires a rebalance of your server-side state, which will affect all members. Technically, it offers less availability, since the cluster can never lose more than N members between rebalances (where N is the number of owners), but can more easily spread load evenly since a given piece of state is available from any node in the cluster.

[valsaraj007]

Thanks pferraro! Yes, I referred that blog.