Cluster configurations

Cluster software vendors support several distinct types of cluster configuration: asymmetric, symmetric, N+1, and N-to-M clusters.

Asymmetric clusters

Asymmetric clusters are asymmetric in the sense that the cluster includes both active and passive servers. An asymmetric 1-to-1 cluster has one active server and one passive server. This is the simplest form of server cluster, but also the most expensive, in effect doubling the number of servers required for a given workload. There is no dropoff in performance after a failover in an asymmetric 1-to-1 cluster, assuming the passive server is identical in capability with the active server.
Figure 1. Basic asymmetric cluster setup
Basic asymmetric cluster setup

Symmetric clusters

Symmetric clusters are termed symmetric because all the servers (also called nodes) in the cluster are active. Although there are no passive servers in a symmetric cluster, the servers typically run different services. No two servers provide the same service for the same data set simultaneously because of the difficulty of maintaining data integrity mentioned earlier.

The net result of a failover in a symmetric cluster is that both services end up running on one node. Note the potential drop in performance if both nodes are not provisioned with enough idle capacity to accommodate running database. For this reason nodes in a symmetric cluster might have to be more powerful than they would be in a non-highly available environment, but no extra nodes are required, in contrast to an asymmetric cluster.

Restriction: You can set up symmetric clusters only with components that can be collocated.

N+1 clusters

An N+1 cluster has N active nodes and only one passive backup node. The lone passive node acts as the backup for all N active nodes. This is an asymmetric cluster, since it has a mix of active and passive nodes, but it requires fewer nodes overall than an asymmetric 1-to-1 cluster configuration for each active node. Each node in an N+1 cluster is configured with the software for all the services supported by the cluster. This allows each node to act as the backup for all the active nodes in the cluster. The benefit is that a failed node, when it comes back online, can take on the role of backup node. This avoids the need for a failback from the former backup node to the repaired node, eliminating the disruption in service caused by a failback event.

Example of an IBM® FileNet P8 N+1 cluster with three nodes:
  • Remote database server node
  • Backup server node

All three nodes are configured to run the database. If the database node fails, a failover to the backup node occurs and the appropriate software is started to take over the role. When the failed node is repaired and brought back up, it becomes the new backup node in the cluster.

N-to-M clusters

An N-to-M cluster has N active nodes and M nodes configured as failover nodes. The M failover nodes can be either active or passive. Like N+1 clusters, an N-to-M cluster can be configured to avoid the need for a failback when the failed node is available for use again, in the case where a passive backup node is desired. If there is no passive node in an N-to-M cluster, a failback is presumably preferred to get the cluster back to its highest performance configuration with all nodes active. The advantage of an N-to-M cluster over an N+1 cluster is that it can accommodate multiple failures, since it has M failover targets that can take over for a failed node.