Exchange supports clustering since version 5.5. This was not fully cluster-aware. Instead it made use of the Windows Cluster's Generic Resource DLL to support its clustering functionality. The first pure cluster-aware version was Exchange 2000.

Tip: What is a Cluster-aware application?
An application that understands the functions or messages generated by the Clustering Software is called a cluster-aware application. Cluster-awareness comes from a Resource DLL which talks to the Resource Monitor of the Clustering Software.

In this article we will discuss and explain the following topics in detail:

• What is Clustering?
• How does Exchange Clustering works?
• Which components make up an Exchange Cluster?
• How to cluster Exchange Server 2000/2003/2007?
• Exchange Clustering requirements, best practices and Limitations

What is Clustering?

A cluster is a set of computers running together to ensure the availability of a service or application to client computers accessing it over the network. There are many clustering technologies available today. Exchange could also be clustered with VERITAS Cluster Server but we will limit our discussion to the Microsoft Clustering Technologies.

Cluster support was introduced with Windows NT Enterprise Edition and continues to be included with the latest Operating Systems. Microsoft provides two clustering options: Microsoft Cluster Server (MSCS) and Network Load Balancing (NLB). MSCS is for failover clustering, whereas NLB is for load balancing the TCP/IP requests coming from the clients over the network. In this article, we will limit our discussion to the MSCS failover technology. We will not discuss anything about the Load Balancing or NLB clusters.

Tip: What is Active and Passive node?
An Active Cluster Node is a node on which at any given time all the resources of an application will be running. A Passive node is waiting for a failure and is responsible to bring the resources online in case the Active Node fails.

There are two types of MSCS clusters: Active/Active and Active/Passive.

Active/Active: In this mode, all the resources of an application will be running on both nodes, serving multiple client requests coming over the network.

Active/Passive: In this mode, one node will be serving client's requests coming over the network and another node waits for the Active Node to go offline to take the resources online. Microsoft recommends active/passive clustering for Exchange. I will discuss this later as to know why this recommendation is given.

How does Exchange Clustering Works?

Clustering software is the base for every application which is to be clustered. Without this base, it is not possible to cluster any application. Microsoft Clustering software understands an application by its type, its resource DLL, and its functionality in the cluster.

Tip: What is a Resource DLL?
A Resource DLL is a component of a cluster-aware application, used by the Clustering software to track the application's availability, as shown in figure below.

Figure A.1 - Monitoring resources via Resource DLL

In the simple figure above, you can see how clustering works. Node A which is Active serves the client's requests coming over the network and reports the status of its resources (Exchange Resources) to the Clustering Software every 20 seconds via the Resource DLL. If anything goes wrong with Node A, the Exchange Resources will be failed over to Node B which is passive in this example. In this example, we are running two nodes. However we could have more nodes depending on the Operating System version.

Tip: What is a Resource Monitor?
A Resource Monitor is a Clustering Software component that requests Resource DLLs to report the application status.

In the above example, client computers are using the Exchange Virtual Server (EVS) IP Address 192.168.1.9 to access Exchange. If anything goes wrong with Node A, the EVS resource including its sub-resources will be failed over to Node B. Clients will be able to access the Exchange Server as usual using the Exchange Virtual Server IP Address 192.168.1.9. Finally the Resource DLL at Node B will be active to monitor the Exchange resources and report the status to the Clustering Software in the same manner it was active on Node A.

Tip: What is an Exchange Virtual Server (EVS)?
An EVS identifies the role of Exchange in a cluster environment. The EVS is generally identified by its name or IP Address as shown in the above figure.

Which components make up an Exchange Cluster?

Operating System Version
Depending on the number of nodes required, you need to select the Operating System version to be used for clustering Exchange. Clustering Software ships with the Operating System Advanced and Datacenter Editions. For node limits, please see the "Exchange Clustering Facts" section in this article.

Network Cards
In a two node clustering, you need at least four NIC cards. One NIC card is required for client computers accessing the EVS over the network. The other NIC is for clustering software to exchange heartbeats.

IP Addresses
The number of IPs required depends on the number of Exchange Virtual Server instances. As an example, if you are planning to install one Exchange Virtual Server in a two node Active/Passive cluster then you need at least 5 IP Addresses:

2 IPs for Public NICs on both nodes

2 IPs for Private NICs on both nodes

1 IP for the Exchange Virtual Server

Client computers will connect using the Exchange Virtual Server IP.

EVS Group
EVS Group is the name given to the cluster group in our example. This is required to club all the Exchange Resources in a group for proper visibility. The EVS will contain all the resources related exclusively to Exchange.

Network Name and EVS IP Address
The EVS Group will contain the Network Name and IP Addresses for the Exchange Virtual Server. The Exchange Virtual Server IP Address or the Network Name is used by the client computers to access Exchange over the network.

Shared Storage Disks
A shared disks for each Exchange Virtual Server instance is required. Each must be accessible by all nodes. Also a separate disk is required by the clustering software for Quorum. Quorum is a shared disk connected to the cluster nodes. This disk is used by the clustering software to determine which node should be active and owner of the cluster. More details are available from: http://support.microsoft.com/kb/818675/en-us