You can cluster Exchange 2000 to get the benefits of clustering: increased reliability and performance. But (there's always a but) there are lots of things you need to think about before you start throwing hardware together. This article discusses these considerations. For more information on clustering Exchange -- and the opportunity to ask questions directly to Exchange expert Scott Schnoll -- listen in to a recent expert webcast...
on clustering Exchange 2000. Click here to tune in for this event.
If you're reading this, I am assuming you're either currently clustering Exchange 2000 or you're contemplating clustering Exchange 2000. Regardless of which camp you're in, there are certain dos and don'ts when it comes to clustering Exchange 2000 with the Windows 2000 Cluster Service. In this two-part article, we'll look at the most important -- and often most overlooked -- aspects of clustering Exchange 2000. This first installment discusses planning for your Exchange cluster and getting your cluster ready for installation. Part two talks about installing and configuring the software to make your cluster a reality.
When Microsoft released Exchange 2000, the product featured substantially enhanced support for the Windows 2000 cluster service. The prior version, Exchange 5.5, could be clustered, but only the setup program was actually cluster-aware. The other components that ran in the cluster used generic resource DLLs to provide cluster features such as failover.
Exchange 2000 also uses a cluster-aware setup program, but it goes further: it installs several components that are cluster-aware, as well. But before you cluster Exchange 2000, you have to know what can and cannot be implemented in an Exchange 2000 cluster. One really good resource for this is Microsoft Knowledge Base Article 259197, which discusses the current status of Exchange 2000 components in a cluster. Even if you have read this before, I recommend reading it again as some of the information has changed.
Microsoft's strongly preferred model for Exchange 2000 is Active/Passive, but to see why, you have to understand what that means. Active/Passive and its companion, Active/Active, refer to the number of Exchange virtual servers (EVSs) -- all of the hardware and software that makes up an Exchange server -- in a cluster, compared with the number of physical nodes in the cluster. While you may already be familiar with these terms, they have special meaning when used in reference to Exchange 2000. In this context, Active/Passive means you have fewer EVSs than physical nodes. So in a two-node cluster, Active/Passive means a single EVS. Active/Active in a two-node cluster would, therefore, mean two or more EVSs.
Microsoft prefers Active/Passive for many reasons. For example, Exchange 2000 requires a minimum amount of contiguous virtual memory. In an Active/Active model, virtual memory is more prone to fragmentation, and after a while you may find it necessary to failover your EVSs and reboot your nodes to resolve virtual memory fragmentation problems. Another reason is that sizing Active/Passive installations is just like sizing a non-clustered, standalone Exchange 2000 server. With the proper hardware, for example, you may be able to get as many as 10,000 concurrent users on an Active/Passive Exchange 2000 cluster. If you go with an Active/Active model, you'll need to limit the number of concurrent users per node to 1,900, and you'll need to keep tabs on the CPU(s) constantly to make sure they do not exceed 40% utilization for more than 10 minutes.
Regardless of which model you choose, there are some limits to which you must adhere. The first is the four-storage group rule. A single Exchange 2000 server can contain only four storage groups. This rule also applies to clustered Exchange 2000 servers, where a physical node in the cluster can contain only four storage groups at any given time. Because Exchange 2000 supports multiple storage groups, and because multiple EVSs can exist on the same physical node, storage groups in a clustered environment need to be planned very carefully. If you failover an EVS with three storage groups to a node that already has an EVS with two storage groups, then one of the storage groups for the EVS being failed over will not instantiate or mount, because four other storage groups will already be running.
Similarly, you must limit the number of EVSs you create in your cluster. In a two-node Active/Passive cluster, you will have a single EVS by definition. However, in a two-node Active/Active cluster, you can create a maximum of four EVSs.
You'll also need to prepare networking information for your cluster. Each physical node in the cluster will be assigned two static IP addresses (public and private IP addresses) and one NetBIOS name. The cluster -- that is, the union of the multiple systems -- also has a separate static IP address and a NetBIOS name (this is the name you use when connecting to a cluster using CluAdmin). Finally, each EVS has a static IP address and a NetBIOS name. In a single-EVS, two-node cluster, this translates to six IP addresses and three NetBIOS names.
In part two of this article, Scott Schnoll discusses the installation and configuration of Windows 2000 and Exchange 2000 on a cluster. Click here to read part two.
Questions about clustering Exchange 2000? Ask Scott.
Microsoft has some great documentation on clustering Exchange 2000, including its deployment white paper, which you can download here.