[rrd-users] How to specify max change in DS:GAUGE ?

Alex van den Bogaerdt alex at ergens.op.het.net
Sun Nov 25 16:06:23 CET 2007

On Sun, Nov 25, 2007 at 03:18:34PM +0100, Kiss Gabor (Bitman) wrote:
> > By the way: Why do you trust 40 and 41, and how do you determine that
> > in case of ( 75, 41, ... ) it is 75 that has to go?
> > 
> > In other words, what happens if 75 happens to be the first measurement
> > in a series when you (or rrdtool) starts to compare?
> Dear Alex,
> Please don't fuss. :-)
> Remember HW anomaly detector can decide if a data point is valid.
> Other algorithms are also conceivable.
> What about this? "Don't drop the first N values just use them to
> compute average and deviation."

I was responding to:
 "like: next value must be within +/-10 of previous value"

If the OP is not careful, he will get wrong results. And per Murphy's
Law, this will happen sooner or later, go undetected and/or cause

Perhaps a cdef using trend and limit will work out nicely. But that
would ignore faulty values at graph time, not at data insertion.

If you know how to ignore those values at insertion time, by all means
educate the Op and me.

Alex van den Bogaerdt

More information about the rrd-users mailing list