Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I need to establish a performace basline for our Terminal Server machines.

i plan to use perfmon , and will have a SCOM server on the site very soon.

The terminals are used as a desktop environment for thin clients.
The users use office, outlook, IE8 for intranet apps (sharepoint portal internal sites)
and a .NET 2.0/3.5 enterprise data driven app.

some of the the servers are virtual hosted by Win2008 hypervisor, some are physical.
all are Windows 2003 SP2 32bit. all are on a SAN storage.

what counters should i gather ?
whats the sample interval ?
leave it always running , or have it run for a a sepcified period?
where can i find what are the acceptable ranges for the counters gatherd?

thanks in advance

share|improve this question

In the past, I've simply done with this SNMP collection and graphing. A 5 minute interval should be just fine; that's the MRTG default standard, for instance.

Unless you've got specific needs, I think the things to care about are simply CPU and Mem utilization, maybe page file utilization, and Queue Length for disks. Network IO may be important, or it may not - depends on your infrastructure and usage patterns. It can't hurt to monitor it, of course; you assumptions can start out correct and end up wrong later.

If you can, also get a graph of number of TS sessions; not sure of the SNMP OID for that one.

Since these are TS machines running normal interactive desktop apps, you really just need to know when performance degrades and how many users were on at that time.

share|improve this answer
im afraid MRTG isnt an option for me here, but i think i get what your saying. still i need to know how to interprate the data. – Menahem May 5 '11 at 14:19
MRTG was just a suggestion. My suggestion is to use an SNMP grapher of any sort, because it makes the collection and interpretation fairly easy. I've never really enjoyed working with Perfmon for long-term performance graphing. – mfinni May 5 '11 at 14:22
OK got it , so say i did the gather & graph, how do i know when a perf counter has a "bad" value and i have a perf problem? – Menahem May 5 '11 at 14:30
100% for CPU or RAM is bad. Average higher than 1 per disk is bad for queue length, generally. Beyond that, you just need to have the numbers to refer to when your users start complaining about performance. There's a whole corpus of knowledge on tuning TS/Citrix servers for good performance (and low latency, for interactive workloads); but every environment and workload and userbase is different. – mfinni May 5 '11 at 14:54

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.