High Availability and Failover 
A J2EE cluster has the following single points of failure (SPOF):
· Database
The other cluster elements (dispatcher and server processes) represent no SPOFs in the sense that a breakdown in one of them would not affect the entire system.
· The dispatcher is stateless, this means that running transactions are not hindered. It can be restarted, or from the outset, two dispatchers can be started.
· If a server process fails, the session is terminated. One way of getting round this problem is to make the session persistent.
For more information about failover at application level see:
Failover for J2EE Web Applications
Failover for RMI-P4 Applications
New Communication Schema
With J2EE Engine 6.20 there were no Central Services, and each cluster element communicated with all other cluster elements. With a large number of cluster elements, system performance was significantly lower, which is why SAP has changed the architecture for 6.30.
The graphic below shows the cluster elements and the communication between them. The single points of failure are shown at the bottom.
You can see that the server processes do not communicate with each other. The advantage of this communication schema is that the number of connections only increases linearly with the number of server processes and not quadratically as in the earlier procedure.
Further Information
You can find the high availability guide for the SAP Web AS here: SAP High Availability
The components relevant for the SAP J2EE Engine are described in SAP J2EE Engine.