Topic
IC4NOTICE: developerWorks Community will be offline May 29-30, 2015 while we upgrade to the latest version of IBM Connections. For more information, read our upgrade FAQ.
4 replies Latest Post - ‏2013-01-08T10:09:09Z by SystemAdmin
SystemAdmin
SystemAdmin
2402 Posts
ACCEPTED ANSWER

Pinned topic Using old nmon data

‏2012-10-17T11:36:32Z |
I'm working with nmon data that is several month old (All already recorded)I'm trying to merge up to a month of data to show some trending over longer periods of time, but run out of resources trying to run that much data. At the most, I've been able to merge about 7 or 8 days worth. Is there a way on the analyzer end to set it up to I guess graph that data at longer intervals instead of the 2 minutes which is pulled into the nmon data?
Updated on 2013-01-08T10:09:09Z at 2013-01-08T10:09:09Z by SystemAdmin
  • nagger
    nagger
    1635 Posts
    ACCEPTED ANSWER

    Re: Using old nmon data

    ‏2012-12-30T17:34:45Z  in response to SystemAdmin
    Hi,
    nmon is really for performance tuning but now you are trying to use it for capacity planning.
    This data over load is the first problem - and it basically a Excel problem.

    The next problem you will find is (even if you fix the merge 30 days) you would have 21600 data points on the one spreadsheet.
    On my larger screen it has 1600 by 1200 pixels for it would be trying to display 13 data point on each pixel !!

    Then the next problem is the days of the week - all would would see in the wildly fluxuating data is the weekdays are busier than the weekends.
    I think you know this already.

    I would decide which is your busy day of the week (say Friday) and then just graph those days (Fridays).

    I hope this helps, Nigel
  • SystemAdmin
    SystemAdmin
    2402 Posts
    ACCEPTED ANSWER

    Re: Using old nmon data

    ‏2012-12-31T11:00:02Z  in response to SystemAdmin
    Here is how I do it
    I have a special version of the nmon2rrd where I have changed snapshot count from 1024 to 40000.
    (+ some rrd graph tuning)

    Then I run this nmon2rrd for the first dataset with execute option which creates the rrd files.
    All datasets after that without execution and run separately these for each of them:

    rrdtool - < rrd_update
    rrdtool - < rrd_top
    rrdtool - < rrd_graph

    (rrd_create would create empty rrd files again, so its ignored)

    Then I get longer, monthly trends updated daily. Unfortunately top process stats will not work for
    obvious reason. See attached file for example.
    • nagger
      nagger
      1635 Posts
      ACCEPTED ANSWER

      Re: Using old nmon data

      ‏2013-01-08T09:31:45Z  in response to SystemAdmin
      I saw a forum entry somewhere but then failed to find it later
      on how you changed to 40000 snapshots.

      Can you let me know the details again - I would like to make that a command line option.

      Cheers, Nigel Griffiths
      • SystemAdmin
        SystemAdmin
        2402 Posts
        ACCEPTED ANSWER

        Re: Using old nmon data

        ‏2013-01-08T10:09:09Z  in response to nagger
        Since I collect monthly data, I force rrd to print day of the month with --x-grid. But
        I think its best to leave it to RRD to decide and not to adjust that along with snapshot counts.

        And with snapshots getopt, I'd like to see "incremental" command line option too which should
        be exclusive to the -x option. Incremental would not delete rrd files and executes all except
        rrd_create. incremental works for all graphs but not for top processes.
        RRD forced to draw day of month instead of long version which made the graph unreadable:
        536c536
        < "%s%s\" --start %d --end %d --width %d --height %d --title \"%s %s %s\" --x-grid DAY:1:
        MONTH:1:DAY:1:604800:%%e ",
        ---
        > "graph %s.gif %s -v \"%s%s\" --start %d --end %d --width %d --height %d --title \"%s %s %s\" ",

        How many snapshots are stored in the rrd:

        574c574
        < fprintf(cfp," RRA:AVERAGE:0.5:1:%d\n",44640);
        ---
        > fprintf(cfp," RRA:AVERAGE:0.5:1:%d\n",snapshots);
        31d seconds backwards:

        1092c1092
        < utc_start = tarray[1]-2678401;
        ---
        > utc_start = tarray[1]-1;

        Hope this helps ;-)

        PS: I still use rrd 1.2.30 since 1.4.x graph creation was **much** slower for me.