This discussion has been locked. The information referenced herein may be inaccurate due to age, software updates, or external references.
You can no longer post new replies to this discussion. If you have a similar question you can start a new discussion in this forum.

Sample (Performance) Data Collection Intervals & Retention

So I was reading the VMan Admin guide the other day (everyone does that, right?) I stumbled across something that I thought was important enough to write a post about so that, when I go searching for it in the future, I'll be able to find it!  It's about data collection intervals.  One of the hardest concepts for folks who are new to network monitoring is the concept that the data collected is not 'real-time'.  (If you are ever sitting a pre-sales engagement with the sales team and one of them says "We can provide real-time monitoring" you have my permission to smack them with a frozen cod)  Real-time suggests that the data is streaming to the NPM or, in our case, the VMan, server.  Not only would the network team hate you for that, but so would the DBAs, storage, server and just about everyone else who wanted to do something with the network other than monitoring.

With that I present a summary of page 89 in the Virtualization Manager Administrator's Guide (v6.1) <drum roll>

  • Performance data is collected every 10 minutes by default
  • Each 10 minute interval is comprised of 20 second chunks of data (or 30 chunks for every 10 minutes)
  • Each 20 second chunk is a single value for that period and can be a Average, Maximum or Last measure.  (Most of the values are averages)

(Here's where it gets a little interesting)

  • The value displayed in VMan is calculated as follows:
    • If an average value, it is the 10 minute average for the collection period.  This means that the value you are seeing is an average of averages!
    • If a peak/max value, it is the single highest value from all of the 20 second intervals collected in that 10 minute period.
      • This means that the peak value may have only occurred once in a 20-second interval in the previous 10 minutes.
    • A 'last' value is simple the last value collected before the query [I'm actually not 100% sure on this one -- anyone else?]

Once all this data is collected it is retained on the following schedule (all values are defaults)

  • Raw data (30 chunks for every 10 minute period) is retained for 14 days   
  • Hourly rollups are retained for 90 days
  • Daily, weekly, monthly and quarterly rollups are retained indefinitely

I certainly hope that this comes in useful for someone else.  I know that for my NPM customers I would spend a fair bit of time explaining to them about data collection and retention periods as it influences everything from the interpretation of reports to alerting.  I expect that our rollout of VMan 6.1 will result in similar discussions.