Clustering Concepts

Because Exchange 2003 leverages the underlying clustering technology found in Windows 2000 Server and Windows Server 2003, the concepts and terminology used to describe clustering Exchange are the same.

To start, a cluster is a combination of two or more servers that act as one. The servers that make up a cluster are called nodes. These nodes work together to ensure that server applications that are installed on the cluster are available at all times.

Servers within a cluster keep in constant contact and can detect failures by checking for a "heartbeat." Similar to the same function in a human body, a server heartbeat is used to determine whether a server node is up and running. If there is no heartbeat, then any work that was to be performed by that particular cluster node is redistributed to other nodes within the cluster.

Clustering was initially introduced in Windows NT 4.0 Server. In terms of functionality and features, clustering has grown alongside the Microsoft Server operating systems. Exchange 2003 can be installed on a cluster running either Windows 2000 Server or Windows Server 2003; therefore, this chapter will concentrate on the new clustering features in Windows Server 2003.

CLUSTER TYPES

You can configure two different types of clusters: either a generic cluster, which is used to scale applications and services across multiple applications, or a "network load balancing" cluster, which can be used to distribute network load across multiple servers. This chapter looks at generic clusters, which is what Exchange 2003 utilizes to scale messaging platforms.

This doesn't mean that you can't successfully install and run Exchange 2003 on a cluster that is running Windows 2000 Server. However, because the underlying clustering technology provides most of the benefits to Exchange, you might want to consider implementing your Exchange cluster on Windows Server 2003 because this platform has several new features that make it a logical choice.

CLUSTERING FEATURES IN WINDOWS 2003 SERVER

For more information on the clustering features in Windows 2003 Server, pick up a copy of the Windows Server 2003 Delta Guide (ISBN 0-7897-2849-4), also published by Sams.

New to this release, Windows Server 2003 provides the ability to create up to an eight-node cluster, where seven of those nodes can be active and one node can be passive. This means that seven nodes within the cluster will be actively supporting each other, available to step in when required to handle processing requests, with one node passive in the background, able to handle requests if the seven active nodes are down (which would occur rarely.)

In terms of providing scalability for Exchange 2003, this 7/1 configuration might not seem like a lot of processing power, but considering that you could use multiprocessor servers with a large amount of memory in each of the nodes, you have the ability to scale to many users.

This scalability really comes into play in one of the most common Exchange clustering scenarios, running a front-end/back-end topology. In this configuration, most Exchange administrators will cluster the critical back-end server components and leave multiple front-end servers to route requests to the back-end cluster. The combination of front-end and clustered back-end servers provides the most reliable configuration and can scale to tens of thousands of users.

You might be saying to yourself, "I can get the same reliability with hardware redundancy." You are right: You could put servers in place with hot spares and work out your own fault-tolerant architecture. The method you choose depends on your budget and requirements. Clustering provides a great measure of reliability but might not cater to common Exchange problems such as a corrupted information store. In this case, a combination of clustering and hardware solutions might be the best fit.

CLUSTERING: NOT FOR LOAD BALANCING

Exchange clustering doesn't provide automatic load balancing. In other words, the nodes in the cluster won't share the workload of handling user Mailboxes. You can manually load-balance the work by evenly distributing users' Mailboxes across the nodes, but you'll have to continually fine-tune that distribution. Because Exchange ultimately stores the actual user Mailbox data in a single location, there's no real way to load-balance back-end servers.

Now that you are excited about the possibilities of clustering Exchange, let's drill down into the details of how it all works, starting with some prerequisites that need to be met before you get started.