[rrd-users] False positives with aberrant behavior detection

Tobias Oetiker tobi at oetiker.ch
Tue Aug 17 08:23:44 CEST 2010

Yesterday Mike Schilli wrote:

> On Mon, 16 Aug 2010, Tobias Oetiker wrote:
> > you may find some inspiration here
> Hmm ... actually, I still have a two questions:
> * My graphs are jagged, due to the sensor's discrete values, delivered in
>   steps of .5.  Is there a way to 'smooth' the graph, so that I get
>   something like splines as input for ABD?


> * Is there a way to change the confidence band from what your document
>   calls "think something like standard deviation" to something twice this
>   size? I've tried many settings of alpha, beta, and gamma, but it seems
>   to be impossible to get rid of the false positives. If I could double
>   the width of the confidence band, however, I'd get exactly what I need,
>   I think.

the confidence band initelf is dynamic and derives its 'width' from
experiance of previous prediction accuracy ...

the best course of action in my experiance is to prepare about 5
cycles worth of hires data and use a little script to visualize hw
behaviour with differenc values ... as I did in my sample ...


> -- Mike
> Mike Schilli
> m at perlmeister.com

Tobi Oetiker, OETIKER+PARTNER AG, Aarweg 15 CH-4600 Olten, Switzerland
http://it.oetiker.ch tobi at oetiker.ch ++41 62 775 9902 / sb: -9900

More information about the rrd-users mailing list