Microsoft Exchange Server - Clustering and High Availability

Clustering and High Availability

Exchange Server Enterprise Edition supports clustering of up to 4 nodes when using Windows 2000 Server, and up to 8 nodes with Windows Server 2003. Exchange Server 2003 also introduced active-active clustering, but for two-node clusters only. In this setup, both servers in the cluster are allowed to be active simultaneously. This is opposed to Exchange's more common active-passive mode in which the failover servers in any cluster node cannot be used at all while their corresponding home servers are active. They must wait, inactive, for the home servers in the node to fail. Subsequent performance issues with active-active mode have led Microsoft to recommend that it should no longer be used. In fact, support for active-active mode clustering has been discontinued with Exchange Server 2007.

Exchange's clustering (active-active or active-passive mode) has been criticized because of its requirement for servers in the cluster nodes to share the same physical data. The clustering in Exchange Server provides redundancy for Exchange Server as an application, but not for Exchange data. In this scenario, the data can be regarded as a single point of failure, despite Microsoft's description of this set up as a "Shared Nothing" model. This void has however been filled by ISV's and storage manufacturers, through "site resilience" solutions, such as geo-clustering and asynchronous data replication. Exchange Server 2007 introduces new cluster terminology and configurations that address the shortcomings of the previous "shared data model".

Exchange Server 2007 provides built-in support for asynchronous replication modeled on SQL Server's "Log shipping" in CCR (Cluster Continuous Replication) clusters, which are built on MSCS MNS (Microsoft Cluster Service—Majority Node Set) clusters, which do not require shared storage. This type of cluster can be inexpensive and deployed in one, or "stretched" across two datacenters for protection against site-wide failures such as natural disasters. The limitation of CCR clusters is the ability to have only two nodes and the third node known as "voter node" or file share witness that prevents "split brain" scenarios, generally hosted as a file share on a Hub Transport Server. The second type of cluster is the traditional clustering that was available in previous versions, and is now being referred to as SCC (Single Copy Cluster). In Exchange Server 2007 deployment of both CCR and SCC clusters has been simplified and improved; the entire cluster install process takes place during Exchange Server installation. LCR or Local Continuous Replication has been referred to as the "poor man's cluster". It is designed to allow for data replication to an alternative drive attached to the same system and is intended to provide protection against local storage failures. It does not protect against the case where the server itself fails.

In November 2007, Microsoft released SP1 for Exchange Server 2007. This service pack includes an additional high-availability feature called SCR (Standby Continuous Replication). Unlike CCR which requires that both servers belong to a Windows cluster, typically residing in the same datacenter, SCR can replicate data to a non-clustered server, located in a separate datacenter.

With Exchange Server 2010, Microsoft introduced the concept of the Database Availability Group (DAG). A DAG contains Mailbox servers that become members of the DAG. Once a Mailbox server is a member of a DAG, the Mailbox Databases on that server can be copied to other members of the DAG. When you add a Mailbox server to a DAG, the Failover Clustering Windows role is installed on the server and all required clustering resources are created.