Advice for clustering

Create cluster members based on your knowledge of the application and on the expected workload. Keep the following considerations in mind:

• Cluster members do not need to reside on the same machine.
• Clients can have inconsistent views of configuration information in the cluster. This can occur when an application server is stopped, started, added, or deleted. The period of inconsistency is short-lived, however. Clients eventually refresh their caches of server information. Application servers that are unchanged during the period of inconsistency remain available.
• In most cases, if you make changes to a cluster, you do not need to restart its cluster members and clients. The changes you made eventually propagate to the cluster members and clients. However, in some cases you need to stop and restart the cluster members of the cluster, for example, when you change the selection policy for the cluster.
• You can make changes to a cluster while it is running. However, incremental changes (such as adding or removing one or two cluster members) have less impact on client performance than wholesale changes.
• It is always best to make changes when few clients and application servers are running.
• You can cluster the initial number of application servers based on the expected load for an application. Later, you can respond to the load on an application by adding or removing server cluster members.
• When a machine becomes unavailable, you do not need to reconfigure the cluster members of other application servers to compensate for any unavailable application servers on that machine. However, when the machine is going to be unavailable for an extended period, you might want to reconfigure the other servers to optimize performance.