Planning Servers

After you’ve planned the general structure of your organization and routing groups, you can plan your servers. The number of servers you need depends on the number of users in the routing group and the services that you plan to provide to those users. As you’ve learned throughout this chapter, you accomplish part of server planning while planning your organization and routing groups. After you have accomplished that, you should have a fairly good idea of the services that each routing group needs to offer and the number of servers that you need to have to offer those services.

Depending on your needs and resources, you are likely to have decided whether to concentrate your services on just a few powerful servers or instead to distribute the services among a larger number of less powerful servers. There really is no guideline for the number of servers that you need or the power of those servers. What is important is that you make a plan. After you make that plan, you can begin to estimate the hardware requirements for your servers.

When estimating the performance of an Exchange server, you need to consider four distinct categories of hardware: disk, processor, memory, and network. The sections that follow discuss each of these categories in turn.

Disk Considerations

Your server needs to have adequate disk space for Windows Server 2003, Exchange Server 2003, directory information, transaction logs, and information stores. The speed at which Exchange Server can access your disks is another important consideration.

SCSI drives are generally faster than IDE drives. Consider using a caching disk controller with a high-speed bus, such as PCI. Adding more drives and drive controllers allows Exchange Server to distribute the workload, reading and writing to multiple drives at the same time. Also consider placing your transaction logs on a separate physical disk so that the logs can be written sequentially, increasing performance.

Real World—Calculating Disk Space

When you plan the amount of disk space your server will need, consider these factors:

• Windows Server 2003, Enterprise Edition (with Internet Information Services, Active Directory, and DNS), takes up about 1.5 GB, depending on the options you install.

• A standard installation of Exchange Server 2003 will take up an additional 180 MB.

• In addition to these figures, you need to factor in the number of user mailboxes and public folders on your server and the amount of space you plan to allow each type of store to consume.

• Transaction logs are relatively small (5 MB each), but you should use a separate drive for them.

Finally, you need to take into account any additional services you need to run on the server, including other major programs such as Microsoft SQL Server and Exchange extensions such as virus and filtering programs.

If you use multiple drives, you might also want to consider a hardware or software-based redundant array of independent disks (RAID), including disk striping with parity (RAID-5) or disk mirroring (RAID-1), to offer some level of fault tolerance. You can configure Windows software-based RAID in the Computer Management snap-in. Computer Management allows you to create a volume set, which is a group of hard disks that the Windows operating system treats like a single hard disk. Yet another option is to implement a form of hardware-based RAID, which can be costly but offers the best performance and fault tolerance available.

Although it might seem tempting to throw as much storage space at the Exchange server as you can, don’t do it. Instead, think about the storage needs over time and the capability of the backup system. If the storage space might exceed the capacity of the backup system, you might need additional servers instead. Many gigabytes of data can accrue on an Exchange server over time. Eventually, the information stores can grow to be too large for the backup system. When defining the storage for a server, ensure that the backup system is adequate to fully back up the information stores, transaction logs, and operating system files. A large information store can take several tapes, and a very long time, to back up on a daily basis. A restore can take several hours. Multiple servers with smaller information stores provide an inherent tolerance to failure: the failure would affect fewer users for a shorter period of time because the restore process is shorter.

As the amount of data on an Exchange server grows, performance can diminish. Exchange Server manages a number of background tasks for the information stores. These tasks take longer to execute when there are more messages in the information store to manage; hence, performance degrades across the server as a whole. One way to keep the mailbox store from growing too large is to configure the server to limit the size of users’ mailboxes. Another way is to configure multiple mailbox stores, which you learn how to do in Chapter 11, “Using Storage Groups.”

Processor Considerations

Using multiple processors significantly increases a server’s performance. Adding a second processor to a server, however, does not double its performance. The processors still share a motherboard, adapters, storage, and other components, and data can face a bottleneck in these components. However, Windows Server 2003 does support symmetric multiprocessing, and Exchange Server 2003 is a multithreaded application. Therefore, multiple pieces of the Exchange system can run simultaneously on different processors within the same system, significantly increasing response time.

Memory Considerations

Memory (RAM) is used to run active processes on a computer. When physical memory is not sufficient, the system supplements it by using a paging file on the computer’s hard disk. Ideally, you should have enough physical memory on a server to avoid excessive use of the paging file. Right now, memory is the cheapest way to increase the performance of any computer. We recommend having at least 512 MB of RAM on any Exchange server as the bare minimum; use 1024 MB from the start if at all possible.

Network Considerations

The network interface cards on your servers should be fast enough to handle traffic coming from and going to clients and other servers. High-speed network adapters, such as those that use a PCI bus, are best. Fast servers can take advantage of multiple network interface cards, providing the ability to host connections to several other clients or servers at the same time. Furthermore, many server platforms allow you to merge network interface cards into a pool; then, should one of the cards in the pool fail, another card takes over. We recommend using high-speed PCI adapters that are capable of bus mastering whenever you can.

Ways to Add Fault Tolerance

Some standard precautions can be taken to ensure that Exchange servers stay on line, even when there are failures. An uninterruptible power supply (UPS) is a common way to ensure that the server does not go off line if the power in the building fails. A UPS can also prevent power surges from damaging the server components.

As we have already mentioned, a server can have multiple hard disks, multiple processors, and multiple network interface cards. These redundant components provide increased performance, load balancing, and failover options, depending on how they are configured. A server can also have dual power supplies, controller cards, and error-correcting RAM. Whenever a server has redundant internal components, it is better able to tolerate faults in those components. Server-class machines typically come with software that is able to monitor the servers’ hardware components from a central management machine.

In addition to establishing redundancy for server components, you can establish redundancy for the server itself. Exchange servers can be configured to take advantage of a shared storage system using clustering. Clustering is a system in which multiple servers are configured in a cluster, so that if one server has a problem, the system fails over to the redundant server.