RRD graphing inconsistiency



  • Hi, I started new RRD graphs with a new install which caused me to realize an inconsistency in the RRD graphing:

    My 1 week, 1 hour average peak is 2.92 Mb/s down on one of my lines
    My 1 month, 1 hour average is 2.09 Mb/s down on the same line

    3 month, 1 day - 647 Kb
    1 year, 1 day - 586 Kb
    4 years, 1 day - 442 Kb

    Similar results on other interfaces. Obviously the peak of the larger windows for a given sample length (say 1 day) should NEVER be lower than the peak of the smaller windows since it's supposed to show the heaviest peak day in the 1 year window, which can't possibly be lower than the heaviest peak day in the smaller 3 month window, for example.


  • Rebel Alliance Developer Netgate

    IIRC, The peak is calculated from the samples in a data set. Those samples are averaged over longer times on the older graphs, so they will go down as the bandwidth averages out.
    It's reporting the peak of the samples on the graph, not a peak overall.

    So if you had a one minute sample at 100Mbit/s and the others at 0, the first graph would be a 100Mbit/s peak since it was seen directly. The next would have averaged that 100Mbit/s out to 100/5=20Mbit/s peak.

    Then the next graph would have averaged that out to 20/12=1.667Mbit/s (12 5-minute samples in an hour).

    And so on.



  • @jimp:

    IIRC, The peak is calculated from the samples in a data set. Those samples are averaged over longer times on the older graphs, so they will go down as the bandwidth averages out.
    It's reporting the peak of the samples on the graph, not a peak overall.

    So if you had a one minute sample at 100Mbit/s and the others at 0, the first graph would be a 100Mbit/s peak since it was seen directly. The next would have averaged that 100Mbit/s out to 100/5=20Mbit/s peak.

    Then the next graph would have averaged that out to 20/12=1.667Mbit/s (12 5-minute samples in an hour).

    And so on.

    But it doesn't behave exactly like that either. Sometimes the longer term graph is higher (even when new so all windows show it). Obviously this isn't your software but there's something in it's graphing that doesn't quite make sense.



  • I think the point was that the window size is the same (1 day) for 3 months/1 year/4 years (at least according to the graphs), so the set of samples for 4 years should include the set of samples for 3 months.



  • @razzfazz:

    I think the point was that the window size is the same (1 day) for 3 months/1 year/4 years (at least according to the graphs), so the set of samples for 4 years should include the set of samples for 3 months.

    Exactly



  • Tangentially related

    The 8 hour graph title indicates 1 minute average.  But he graph granularity is 5 minutes.  Doesn't that really make it a 5 minute average?  Average of five one minute averages.  Is that not the same as a five minute average?

    Thus the title is a little misleading in that it seems to indicate the granularity of the graph.  But looking closer at the time scale reveals the graph to have a 5 minute granularity.

    http://forum.pfsense.org/index.php/topic,58158.0.html



  • That would certainly explain the behavior.



  • @NOYB:

    Tangentially related

    The 8 hour graph title indicates 1 minute average.  But he graph granularity is 5 minutes.  Doesn't that really make it a 5 minute average?  Average of five one minute averages.  Is that not the same as a five minute average?

    Thus the title is a little misleading in that it seems to indicate the granularity of the graph.  But looking closer at the time scale reveals the graph to have a 5 minute granularity.

    http://forum.pfsense.org/index.php/topic,58158.0.html

    Ah that's getting closer to what I'm seeing.


Log in to reply