Albert Einstein reportedly had a sign on his wall that read "Not everything that counts can be counted, and not everything that can be counted counts." The second half of this admirable sentiment isn't true when it comes to Exchange Server design and deployment. Typically, the goal when designing a server is to provide a certain level of service quality or performance (as expressed by a service level agreement--SLA), so anything that we can count to help us objectively measure the level of server performance can be useful.
Microsoft's Exchange Server 2003 Tools and Updates site includes three tools that can help measure server performance, tuning, and scalability. You can use these tools together to test various aspects of your Exchange server design.
The first tool is Load Simulator (LoadSim) 2003. Good old LoadSim has been around for a while, but the most recent version implements the relatively new Messaging API (MAPI) Messaging Benchmark 3 (MMB-3) and can simulate loads generated by Outlook 2003 operating in Exchange cached mode. MMB provides a standardized way to test server performance by making a fixed set of MAPI requests against the server. By tuning the number and type of requests, you can make LoadSim approximate various kinds of messaging loads, from heavy to light. LoadSim simulates only MAPI clients, though; if you're using Internet-protocol clients, you'll need another tool.
That tool is Exchange Stress and Performance (ESP) 2003 (sometimes known as Medusa). What LoadSim does for MAPI clients, ESP does for IMAP, POP, and Outlook Web Access (OWA) clients. ESP gives you a range of configuration options to control which specific protocols to test and what kind of load to present to your server.
The third tool is Jetstress. Whereas LoadSim and ESP simulate client behavior, Jetstress stresses the stuffing out of your disk subsystem by making I/O requests that follow the patterns typical of production Exchange servers. Of course, you still need to use Performance Monitor or a similar tool to watch the parameters you're interested in (e.g., average disk queue length) and see what effect Jetstress is actually having.
ESP and Jetstress work on Exchange Server 2003 and Exchange 2000 Server; LoadSim works on these systems and on Exchange Server 5.5. Each tool comes with documentation that explains how to use it and what it can measure (for example, you probably won't want to use Jetstress to measure disk performance on a test server whose hardware is significantly different from your production equipment's hardware).