Planning Your Simulation

When you are planning your simulation, use the following questions to develop a scenario that will produce reliable results.

• What is the goal? Is it to find an adequate server sizing? To determine your network bandwidth needs? To determine what tuning adjustments can squeeze the most performance out of a given server?
• What is your network topology? Make sure your experiment uses a network that is similar to the one you use in real life. Consider using line speed simulators to reproduce the bandwidth of your connections. Bandwidths between clients and servers are often different than bandwidths between servers and servers.
• Is your topology meaningful? Single-server tests are interesting, but they are usually not the most realistic. Servers usually communicate with other servers as well as with clients. Are your servers also domain controllers? Do your servers run other services in addition to Microsoft Exchange Server? Check your hardware and software configurations carefully. For example, if you forgot to set parity on the RAID disk array, this changes the performance.
• What are the server roles? Some servers support end-user mailboxes directly. Some servers run connectors, and others support public folders.
• What are the server hardware configurations? In addition to the obvious things like CPU type and speed, memory size, disk type and speed, network interface type and speed take into account the system buses, processor cache sizes (L2 cache), and disk controller caches. All of these elements affect server performance.
• How do your users behave? Few people behave in a manner that is identical to the behaviors of the canonical users defined in the Load Simulator profiles. If your company uses a lot of e-mail and few public folders, you would set up the experiment differently than if your company uses more public-folder-based forms. If you already have some users running on Microsoft Exchange Server, you can study the Microsoft Exchange performance counters using Performance Monitor to learn more about your existing (real world) load. There are many aspects of user behavior in addition to messages users send and receive such as deleting mail, moving mail to other folders, rereading, forwarding, and replying to messages. Each client module has a somewhat different set of tasks. Review them to see what information you will need.