Averages: Business / non-business hours RRS feed

  • Question

  • Hello,

    We will be using MAP to find the loads of a number of servers. In turn, these values will be used when virtualizing servers. The scan will run for two weeks. However, I am somewhat worried that the CPU and RAM data will be incorrect.

    The servers are only used during business hours, and the FAQ says that "CPU and Memory utilization figures are computed as averages". If we could run the scan only during business hours, it might find that CPU utilization is 60%. However, since the scan will also include off-peak hours when CPU utilization is basically 0%, we will get a lower average value (about 8/24 * 60% = 20%).

    Now, if I try to virtualize four such servers on one host, the total load "should be" about 4 * 20% = 80%. However, since each server is actually used to 60%, the total actual load *during business hours* would be 4 * 60% = 240%... which would be a Bad Thing™, since servers would be very slow.

    Am I wrong? Does MAP take this into consideration when producing averages? Does MAP take this into consideration when suggesting how many and which measured servers can be virtualized (consolidation recommendation)? Can I use the loads and manually puzzle them together, and still feel confident that the host will not be too burdened?

    Am I right? If so, how would I best go about working around the problem? Can I "handicap" the host CPU to run at only 30% of its actual capacity when I use the MAP consolidation recommendation?

    I'm very glad this forum exists. :-)


    Kind regards,


    Monday, October 18, 2010 1:33 PM

All replies

  • The MAP tool is basically looking for peaks in utilization.  The most effective way to use it (if you already understand your usage patterns) is to start the tool & let it gather data during the time of highest utilization.

    Running the tool for a long period of time (like a week or a months) is most valuable when you don't know when your peaks occur.  If you already have a good idea of when peaks happen, just get the usage data from the time of your peak (at least 30 minutes worth of data) and use that for your placmenent calculations.

    Friday, October 22, 2010 10:30 PM