Dpinger settings and % loss calc



  • Please educate me.

    Parameters:
    Probe Interval: 15000 (15 seconds)
    Loss Interval: 1250 (1.25 seconds) (default)
    Time Period: 60000 (60 seconds; 1 minute)

    With those parameters, and my probably flawed understanding.  I would expect there to be 4 probes per minute, and for the calculations to be based on the last 4 probes.  Yes?/No?  If not please educate me.

    Now IF that is correct I would also expect that the percent loss reported by dpinger could only have values of 0, 25, 50, 75, or 100 percent (0/4, 1/4, 2/4, 3/4, 4/4).

    But in the RRD quality 1 minute graph and data I see max loss % of other values.  e.g. 41.9 %.

    Hope I'm not too inept for someone to be able to educate me.

    Thanks.



  • With the parameters you list, dpinger itself will report based upon the last 3 or 4 probes. Loss values of 100, 75, 66, 50, 33, 25, 0. These are the values that you would expect to see in the widget.

    Rrdtool is a different discussion. It does consolidation of the values fed to it. Search for "rrdtool consolidation" or "rrdtool normalization" for more information.



  • re: dpinger
    I understand that is the result.  But it is not what I would expect.  With a 15 second probe interval and a 1 minute averaging time.  My expectation is for 4 probes to be averaged.  Not 3 sometimes and 4 other times.  Inconsistent.

    The way I would go about achieving that would be to calculate the number of probes to be used from the probe interval and averaging time.
    60000 / 15000 = 4

    So always going to use the last 4 probe results.  So long as all 4 of the probe results are within the last averaging period plus one probe interval.
    60000 + 15000 = 75000 (75 seconds).
    EDIT: Or to tighten that up more use the loss interval in place of probe interval.

    If there are not 4 probes within that time (75 seconds) then no data point exists.

    re: rrdtool consolidation
    I'm aware that it does this to reduce the data set for the longer time period graphs.  But does it consolidate the data in the 1 minute graph too.  I wouldn't think so since there is only 1 data point per minute to begin with.  There should be no further consolidation possible.  Should there?

    The rrdtool is capturing 1 dpinger data point per minute.  That would seem to fit nicely into the 8 hour 1 minute graph without any further consolidation possible.



  • With your configuration, there are 4 slots in the table. Consider the situation in which an echo request has been sent, no response has yet been revieved, and the elapsed time is less than the loss interval. In this situation, only 3 of the slots can be used for the calculation. How often this happens in practice depends upon the round trip time and the relative timing of the probes vs report requests.

    @NOYB:

    I understand that is the result.  But it is not what I would expect.  With a 15 second probe interval and a 1 minute averaging time.  My expectation is for 4 probes to be averaged.  Not 3 sometimes and 4 other times.



  • I understand that.  What I'm saying is that it results in bouncing between n probes and n-1 probes.  Inconsistent.  Believe what I outlined would be more consistent because the data set population being used will always be the same size and thus representing a more consistent time duration.



  • Using a consistent number of probes would require either delaying the report until the pending probes were resolved, or ignoring completed probes that are within the time window. Neither of these is desirable. The problem becomes much more acute when you use a configuration with higher accuracy, where many more probes can be pending.

    The dpinger defaults are designed to provide 1% accuracy. I recommend staying with the default configuration unless you have a strong reason to change it, such as high round trip times.

    @NOYB:

    I understand that.  What I'm saying is that it results in bouncing between n probes and n-1 probes.  Inconsistent.  Believe what I outlined would be more consistent because the data set population being used will always be the same size and thus representing a more consistent time duration.



  • Here is a link that provides a reasonable explanation of how rrdtool processes data. The section entitled "Normalize Interval" is the most applicable to your question as to why you see 41.9%. Enjoy.

    http://rrdtool.vandenbogaerdt.nl/process.php