How many performance counters is too many?

  • I am trying to establish a baseline on a new Sqlserver and am running performance monitor from my local machine across the network. I currently have around 50 counters I am measuring (mainly disk counters per partition, which total 35 for just disk measurements). Is this too many counters to monitor in terms of bandwidth or potential load on the SQL box/network? I am doing this in real-time to get a good feel of whats relevant, then will log for a week or so during the normal business hours to establish a baseline. But I would like to keep this running real-time pretty much all the time to keep a proactive eye on things. Any problems with this I should address?

    *edit*

    I should add the SQL box is currently running wayyyyy below it's capacity as is the network. So this monitoring doesn't appear to be having any negative effects. But I am new to monitoring on this level and just want to make sure.

    *double edit*

    Polling is every 30 seconds for 8 hours in my real-time view.

  • If you're worried about it, start small and add counters as you need them.

    - Gus "GSquared", RSVP, OODA, MAP, NMVP, FAQ, SAT, SQL, DNA, RNA, UOI, IOU, AM, PM, AD, BC, BCE, USA, UN, CF, ROFL, LOL, ETC
    Property of The Thread

    "Nobody knows the age of the human race, but everyone agrees it's old enough to know better." - Anon

  • GSquared (12/8/2010)


    If you're worried about it, start small and add counters as you need them.

    Which is what I have been doing. I started with the basics then added the ones that were relevant to my overall performance. But now that I am at 50, I am wondering if thats too many. How many have you used or heard of people using? From my research it seems the load from performance monitoring on this level is minimal, but I just wanted some real world feedback.

  • I've attached my counters for any feedback as well.

  • Craig,

    You should have no worries with a polling interval of 30 seconds, commercial products dwarf that load.

    The next step is determining which counters fluctuate enough to be relevant (for constant monitoring) and which ones provide information useful in diagnosing root cause. Buffer Cache Hit Ratio may never fall below 98 percent, it can be useful, however, in many cases it isn't beyond establishing a baseline and determining "initial" system health. Page Life Expectancy seems to be more sensitive towards memory pressure. After a few months staring at the same graph all day you'll discover which ones add value for your systems.

    Dan

  • drawlings (12/8/2010)


    Craig,

    You should have no worries with a polling interval of 30 seconds, commercial products dwarf that load.

    The next step is determining which counters fluctuate enough to be relevant (for constant monitoring) and which ones provide information useful in diagnosing root cause. Buffer Cache Hit Ratio may never fall below 98 percent, it can be useful, however, in many cases it isn't beyond establishing a baseline and determining "initial" system health. Page Life Expectancy seems to be more sensitive towards memory pressure. After a few months staring at the same graph all day you'll discover which ones add value for your systems.

    Dan

    Thanks that's what I was looking for. I appreciate the feedback. I am still fine-tuning my counters for this environment, but feel what I have now is a good base.

    *edit*

    I know the disk access read/write times in my attached screenshot are high. That's because I charted the maximum values on that one. My averages are well within reason.

Viewing 6 posts - 1 through 5 (of 5 total)

You must be logged in to reply to this topic. Login to reply