Guidelines for Deploying Exchange 2000 Clusters

You've been asking, so let's talk about Exchange 2000 clustering. Clustering has been a hot topic the past several months as Microsoft works to provide official support for Windows 2000 Datacenter Server and, more specifically, four-node clustering. Microsoft's testing for Exchange 2000 clustering revealed a few important changes in support guidelines and configurations.

With more robust support for Exchange clustering comes increased complexity of design, deployment, and operations. At Exchange 2000's initial release, Microsoft had planned to support all configurations and modes of operation for Exchange 2000 clusters. In addition, Exchange 2000 Service Pack 1 (SP1)—due out in a few weeks—also provides support for Datacenter. Datacenter's significance is its support for four-node Microsoft Cluster service clusters (whereas Win2K Advanced Server supports only two-node clusters). For Exchange 2000, four-node support offers outstanding opportunities for high availability and server consolidation.

Although various cluster deployment configurations and options are possible, Microsoft has recognized the need to limit Exchange 2000 cluster configurations to ensure availability and supportability. Architecturally, Exchange 2000 requires significant amounts of virtual memory allocation for Exchange virtual servers, storage groups (SGs), and user connections, resulting in some specific support limitations and guidelines for deploying Exchange 2000 clusters. Consequently, most deployments will probably select N+1 cluster failover configurations. N+1 refers to N nodes with an additional (+1) node running on standby; virtual servers will fail over to this node during a failure. Active/active configurations (Exchange virtual servers running on all nodes) are supported only on two-node clusters with a concurrent connection limit of 1000 users per node (this limit applies to all protocols). For all other cluster configurations (i.e., three-node and four-node clusters), only N+1 failover configurations are supported. N+1 cluster configurations, however, have no defined connection limit.

The number of virtual servers and SGs that cluster configurations support are also limited. For N+1 clusters, the limit is one virtual server and four SGs per active node. For active/active two-node configurations, the limit is one virtual server and two SGs per active node (in addition to the 1000-connection limit).

These node-configuration support limits might not be as desirable as having no limits, but they're designed to ensure maximum availability for Exchange 2000 clusters. The last thing you want is clustering that increases your downtime. Microsoft has taken some prudent steps with Exchange 2000 clustering to ensure you have a more reliable Exchange deployment should you choose to deploy clusters.