Performance troubleshooting basics

re: Performance troubleshooting basics

Rick — Thu, 29 Sep 2005 23:20:26 GMT

if RPC Average Latency is a problem when it's greater than 50ms, then why is MOM 2005 Exchange Management pack set to alert at >200ms??

re: Performance troubleshooting basics

Nicole — Fri, 30 Sep 2005 03:56:02 GMT

The MOM Pack has a number of different thresholds than the performance guidelines. MOM thresholds are based on relatively small samples times, and lower thresholds would give many false positives. When I blog about 50ms thresholds, I am expecting these latencies to be sustained over a period of time (eg, 30 minutes or more). There are many transient reasons for counters to spike occasionally, and spikes are not always an indication of a performance problem.

Excellent Blog and Summary

Frank Plawetzki — Fri, 30 Sep 2005 10:05:54 GMT

Nicole, excellent work 8)
Thanks for this great guideline

re: Performance troubleshooting basics

Adam Gates — Fri, 30 Sep 2005 21:56:05 GMT

Don't forget the App and Sys logs.
I don't know how many times I have been called to a site and found the App and Sys logs FULL of Disk Buffer Overflow errors, read write failures, and latency issues.
FIRST, check the logs.
My brother is mowing the lawn and the mower dies. HE wants to by a new mower. I check the gas tank... low and behold add gas and it works.
Keep it simple!

re: Performance troubleshooting basics

Blake Handler — Sun, 02 Oct 2005 19:22:42 GMT

Thanks for the article -- (of course) it can not cover everything. . .but it's a great outline for us to follow. I'm sure I'll be "borrowing" this information!

Also looking forward to your new "EXPTA" utility!

re: Performance troubleshooting basics

Kevin — Tue, 04 Oct 2005 16:35:34 GMT

So, if Microsoft sees 0.20 operations per second per user on the server, what would be the upper threshold for this value? Is 1.0 too high? Let's say you came across a server with 500 users and 2 operations per second per user...

re: Performance troubleshooting basics

Nicole — Tue, 04 Oct 2005 20:24:43 GMT

First, let me thank Adam for pointing out that the App and Sys logs are a great place to start to look for problems.

In reply to Kevin's question: there isn't a hard limit on how many operations per second per user is too much. It's just a question of how much hardware you want to buy, and how many applications the users want to run that access the server. If I saw a server with 500 users and 2 ops per second per user, I'd pull out ExMon to identify if the load was caused by a few users. I've quite often found certain apps to be capable of creating extremely high load. When even one or two users run some applications, it can have a noticeable effect on the server. Occasionally we've found an application was incorrectly configured and were able to reduce the load. On the other hand, sometimes everyone in the organization has multiple applications accessing the server, and the applications are important to their business - in this case, we just make sure the server is beefy enough to handle the load. There's just no single simple answer here. Is there enough interest in this for another blog topic?

re: Performance troubleshooting basics

Kevin — Thu, 13 Oct 2005 22:58:36 GMT

I'm definltey interested in any topics around performance and Exchange, the more information I can extract from the Blog team the better!

We do performance logging for Exchange and I know how to look for the big issues but any articles (like this one) that go into steps you would actually take are much appreciated.