Exchanging Problems for Performance

In this issue, we offer guidelines for improving overall Exchange performance, and then focus more closely on disk performance and storage. Future articles will explore utilization of kernel memory and virtual memory performance of the Exchange store.

Exchange Monitoring Part 1: Using MIC Metrics to Boost Overall Performance

As we have discussed in previous issues, any performance tuning exercise should begin with the basics for the operating system: MIC, or Memory first, followed by I/O, then CPU. For complete coverage of Windows server monitoring basics, refer to our August 2005 edition. We'll start with a quick refresher and some particulars to look for on Exchange servers:

Memory: We recommend that you keep your page rate under 1000 pages per second (Exchange servers are expected to have a higher page rate than other systems), and make sure at least 50 MB of memory is available to the Exchange server.

I/O: Try to keep disk utilization under 50% on a sustained basis, and be on the lookout for disk queue lengths greater than two (2).

CPU: Average system CPU percentage greater than 90%, or processor utilization greater than 75%, while the processor queue is greater than two (2), indicates processor saturation and possibly degraded performance.

It almost goes without saying, but any unnecessary software installed on your server can eat up system resources—get rid of it.

It may be appropriate to spread Exchange tasks over multiple servers: for instance, if one machine performs a lot of distribution list expansions, you can reduce its CPU utilization by offloading this task to a dedicated server.

And finally, watch for excessive Exchange usage (in terms of messages and bytes received and sent) by particular Exchange users, domains, servers, etc. You may be able to curb some excess, and re-distribute the rest. Which leads us to:

The Care and Feeding of Disks

Disk space may be cheap, but it's not limitless. Unnecessary disk usage can degrade your disk read/write time. High disk read time (greater than .01 seconds) or write time (greater than .05 seconds) will cause degradation of the page file disk and affect disk performance. Low disk space (less than 20% free) may prevent the creation of new files and cause application failure. To remedy this:

		Enable caching on the array controller and increase log buffers.
		Enforce size limits for both messages and mailboxes. As a general guideline, mailboxes or folders containing more than 500 mail messages, or taking up more than 5MB, may indicate wasteful use of space. Your users will groan, but ask them to delete (or delete for them) any unneeded messages in large folders or mailboxes. Going forward, impose a storage limit on folders and mailboxes, or on the folder or mailbox store.
		Periodically review your disks to see if you can free up space in the file system by archiving or removing files all together.

As Always, Your Mileage May Vary

These are rules of thumb, starting points for developing a clear understanding of your own Exchange workloads. If you monitor these metrics consistently, you will learn exactly what mileage to expect given your particular situation, and be able to focus your energies on the issues that warrant it.

Learn more ...

Next month, Exchange Monitoring Part 2: The Use and Abuse of Kernel Memory