Profiling Exchange performance 1

[primate]

Hi,

I'm currently testing Anti-virus products. I want to be able to get some measure of the impact Exchange aware products have on the performance of an Exchange server. I have Exchange setup in a virtual environment for testing.

I've run the Exchange Loadgen utility but I am not sure that the information it provides is particularly useful, I expected to get some performance metrics at the end of the process but just got value for the number of tasks completed. Conceivably I could run the Loadgen utility again but add some performance counters to the Exchange system, however what would be the best counters to use to profile the effect that AV would have on the database? Things like cache hits and misses shouldn't be affected I would have thought which are the kinds of counters I'd use normally to monitor the health of a database.

Any thoughts appreciated,

TIA.

 

[ShackDaddy]

Testing AV and Anti-Spam products shouldn't be about concern for the health of the database(store), but for the health of the server's ability to process traffic efficiently while keeping its memory and processor optimized.

I've done Exchange server audits before, including performance monitoring, and I've listed my most-used counters here:

http://www.shackelfordconsulting.com/2007/03/exchange-2003-performance-counters.html

And here are some other articles that I've found useful for interpreting counters:

Performance TroubleShooting Basics:

http://msexchangeteam.com/archive/2005/09/28/411674.aspx

RPC Performance Counters Indicate Poor Performance:

http://technet.microsoft.com/en-us/library/d97e8f41-97b8-4858-b51c-bcaf48f48607.aspx

Tuning Exchange 2003 Performance:

http://technet.microsoft.com/en-us/library/d97e8f41-97b8-4858-b51c-bcaf48f48607.aspx

Finetuning Your Exchange Server:

http://gethelp.devx.com/techtips/exo_pro/10min/10min1299.asp

Perf Counters for Users, Logons & Connections:

http://msexchangeteam.com/archive/2006/09/11/428871.aspx

ShackDaddy
Shackelford Consulting

 

[ShackDaddy]

And to be more specific, one of the most important things to look at is the speed at which mail is pushed through the AV and into the local delivery queue. That's a place where you are likely to get latency. Unfortunately, that's not an easy thing to get stats on, unless the AV program includes its own counters.

Best thing for that might be some real-time testing to see how quickly that AV processes 100 real emails. Hard to do in a test environment. I would look and see if a trade magazine like SC has done a comparative study that included these sorts of throughput tests.

ShackDaddy
Shackelford Consulting

 

[primate]

Thanks for your advice Shack, I'll have a read through those articles. The reason I was focusing on database performance was that I thought the AV API hooked into the database? So it seemed logical to see how quickly messages were processed into the database or how big the queue got for example.

 

[ShackDaddy]

Well, AV scanning happens in two places. It can happen on the mailstore itself, but usually when you set up a mail-specific AV, it inserts itself into the inbound mailflow circuit prior to actually hitting the mail store. So it's actually closer to the IIS SMTP engine than it is the Exchange IS subsystem. That's why, with some products, when you shut down the AV service, mailflow stops, since it has become part of the SMTP pipe and not just a disengaged agent looking at items flowing past.

ShackDaddy
Shackelford Consulting

 

[primate]

OK cool, thanks for clarifying.