Scaling Microsoft Exchange 2000: Create and Optimize High-Performance Exchange Messaging Systems

Performance testing for a product such as Exchange 2000 is focused on two distinct evaluation criteria. One is to estimate the capacity of a given server, storage subsystem, and overall combination of hardware and software to handle a certain load with a satisfactory response time. The other is to evaluate the features of the product being tested that you believe your environment will exercise most and to determine what the benchmark or load simulation results reveal about those features, how the proposed solution reacts to the stress testing, and how it can evolve in the future. This evolution study is frequently referred to as "capacity planning."
Whatever tests are carried out, they will always occur in a closed, sanitized environment. The best strategy is first to understand the load generated and the response of the server. Second, when you begin your deployment, closely monitor the performance of your servers, observe their sweet spots, and note how far or close you are from the benchmark environment. You can then proceed to some corrective actions (e.g., more disks, more L2 cache) to arrive at the solution that matches your expectations.
For application service providers, and Internet service providers, it is not practicable to qualify a scalable solution for 5 million users. Often, the benchmark or load characterization has to focus on a representative part of the target environment. It will require skill to extrapolate the results into the target solution and decide whether the results...