Error calculation to estimate error introduced by consolidation

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Error calculation to estimate error introduced by consolidation

Andreas Schuldei
Hi, 

I am in the process to calculate the error introduced by consolidating the data. (I assume my measurements are accurate, I only want to estimate the aggregation error.) Currently I use the MIN and MAX values of my data as the absolut error of my measurements. But then i am not really sure if that is not grossly overestimating the error - AVG is average for a reason, after all. 

Did anyone do some thinking about this? How do i do reasonable error estimation, given that all MIN, MAX and AVG are actual, "accurate" values in the given time frame?

_______________________________________________
rrd-users mailing list
[hidden email]
https://lists.oetiker.ch/cgi-bin/listinfo/rrd-users
Reply | Threaded
Open this post in threaded view
|

Re: Error calculation to estimate error introduced by consolidation

Alex van den Bogaerdt-5
> Hi,
>
> I am in the process to calculate the error introduced by consolidating the
> data. (I assume my measurements are accurate, I only want to estimate the
> aggregation error.) Currently I use the MIN and MAX values of my data as
> the absolut error of my measurements. But then i am not really sure if
> that
> is not grossly overestimating the error - AVG is average for a reason,
> after all.

Remember you're not dealing with values, you are dealing with rates over
time. In other words, don't just look at the Y-axis, you should be looking
at Y-axis multiplied by X-axis.

Make sure you understand how rates are normalized. By the way, perhaps you
are thinking of this part when you say "the error introduced by
consolidating the data" ?

Make sure you understand how PDPs and CDPs are processed. Then it will
become apparent why you should not use MIN and MAX to estimate errors.

Assuming there are no bugs, the only error should come from rounding in
the math done by RRDtool. My guess is that most of them will cancel each
other out.

Please share your findings.

regards,
Alex


_______________________________________________
rrd-users mailing list
[hidden email]
https://lists.oetiker.ch/cgi-bin/listinfo/rrd-users