LOCATION: Zürich

SPEAKER: Bela Ban   COMPANY: JBoss
SLIDES: 040809_12JBossCache.pdf

TreeCache is essentially a tree which provides for its contents to be replicated between address spaces. It supports J2EE transactions (javax.transaction.Transaction), which allows a user to bundle modifications to the tree in a transaction, and to replicate (at TX commit time) or abort (at TX rollback time) all modifications as one unit of work. Alternatively, when transaction support is disabled, replication takes place after each modification.

Replication can be synchronous (blocking the caller until all nodes in the cluster have applied the modification), or asynchronous (replicating the modification in the background).

The TreeCache is used for clustering in the JBoss appserver, but can also be used standalone.

TreeCacheAop is a subclass of TreeCache that accepts regular POJOs (Plain Old Java Objects) as contents. Any POJO added to the tree under a given key will be transparently replicated between trees, and access to the POJO can be transactional.

POJOs do not need to be serializable, because TreeCacheAop uses a combination of reflection and dynamic AOP to (a) determine the structure of a POJO and (b) detect when any fields in a POJO have been changed. Point

1. allows for dynamic mapping of any POJO onto the underlying TreeCache, point
2. allows (through field interception) to record all modifications to a POJO during a transaction.

1. whether an object has been modified or not and,
2. if yes, what fields have been modified.

This allows us to prevent unnecessary replication (point 1) and to replicate only the fields that have changed (point 2), resulting in less network traffic due to unnecessarily replicating entire objects using serialization.

The TreeCache/TreeCacheAop combo will be used in JBoss for HTTP session replication, Stateful Session Bean replication, replicated entity bean cache and the replicated JNDI.