July 2, 2002 at 5:56 am
Hey,
Just a quick question. I usually run perfmon from a server that I connect to using Terminal Services. Basically I use this server like a desktop...I was running a perfmon on CPU usage. I copied the perfmon profile to my desktop and ran it from there. I noticed that the counters were giving me various results. So I started them both within a couple seconds of one another.
I noticed that although they the counter would spike at the same time, I would get different results. In some cases, the %total processor time would vary by upto 20%! And it was not consistent. Sometimes the server would show the higher reading, and othertimes the desktop would show a higher reading. I also noticed smaller spikes for processor queue length. You would see a small spike of say 5 or 6 and nothing on the other perfmon session!
Got me thinking, which one am I to believe? Both machines are regular machines. Nothing to do with SQL Server. The server is a dev machine and the desktop only has terminal services. Both run Win2k.
I'm baffled!!!
Clive Strong
July 9, 2002 at 3:01 pm
Are you sure you're not monitoring your own computer after placing it on your desktop? I suppose also there will be some overhead when running it on the server vs not.
July 9, 2002 at 10:09 pm
Interesting, I'll try to remember to look tomorrow. I rarely run perfmon on the server itself.
Andy
July 12, 2002 at 10:42 pm
The also could be some sort of time division difference that could be the cause. I am not sure as I have not tested, but I believe the data is received in increments and not true realtime otherwise the server would get real busy. If there is a difference in the time division as when started it could potentially return 2 different readings which are from different times in utilization. Look at CPU in Task Manger and see if either matches. This is just a thought and is not tested but I believe that is what is the situation.
"Don't roll your eyes at me. I will tape them in place." (Teacher on Boston Public)
July 13, 2002 at 7:49 am
I want to say while searching for something else I saw an article about not running Performance Monitor through Terminal Services. It was either in the Microsoft Knowledge Base or at a Citrix site. I'll see if I can find it.
K. Brian Kelley
http://www.sqlservercentral.com/columnists/bkelley/
K. Brian Kelley
@kbriankelley
July 13, 2002 at 8:27 pm
It wasn't what I thought. The article refers to counters not appearing correctly in NT 4 prior to SP 6a. The problem is if the chart file is viewed at a system which doesn't have the same type of counters as the system monitored. The example given in the article is if the monitoring has ASP counters but the system where the chart is viewed on doesn't have them. Then all values will be 0. For those still supporting NT 4 prior to SP 6a, here is the relevant article:
http://support.microsoft.com/default.aspx?scid=kb;en-us;Q235519
K. Brian Kelley
http://www.sqlservercentral.com/columnists/bkelley/
K. Brian Kelley
@kbriankelley
July 15, 2002 at 2:33 am
We have observed a similar effect monitoring two clustered servers. When they are both running on the same box (cos 1 has switched over to the other box) the processor utilization stats for the 2 are generally the same but the processor queue length varies (measured through the virutal servers but actually measuring the same box). I think this is because the queue length is an instantaneous figure sent by the server. As it can vary enourmously in fractions of a second it sends a different 'current' value. If you click on explain on some of the counters when adding them, at the end of many of them, it says how they are worked out, & it makes a difference to what you see (& interperet). For queue lenght it says 'This counter displays the last observed value only; it is not an average.' Hope this helps. P.s. I'm not an expert.
Viewing 7 posts - 1 through 6 (of 6 total)
You must be logged in to reply to this topic. Login to reply