[rrd-users] values being averaged.
linux at thehobsons.co.uk
Fri Jan 18 12:04:57 CET 2008
At 23:30 +0100 17/1/08, Fabien Wernli wrote:
>On Thu, Jan 17, 2008 at 12:57:08PM -0500, Shriver, Daniel wrote:
> > "GAUGE does not save the rate of change. It saves the actual value
>> itself. There are no divisions or calculations. Note though, GAUGE,
> > like all other types IS normalized. Thus if a value comes in at an
>> off interval timing (such as 5 min 1 sec after the previous value
>> when the interval is set as 5 min exactly) the value graphed will
>> be an interpolated estimate of what the value should be at 5 min"
>I guess that clears it up quite well.
Almost, but it's not the interval since the last update, it's the
relationship between update times and step periods. So how about
these snippets to go in suitable places :
"Regardless of timezone, all storage intervals are an integer
multiple of 'step' since unix epoch (midnight, 1st Jan 1970*). Unless
updates are submitted for exactly these fixed periods, then the data
(irrespective of data type or consolidation function) is normalised
to fit them. Eg, with 5 minute steps, updates submitted at 10:05:10,
10:10:05, 10:15:05 would be normalised so that 290s (5mins - 10s)
worth of the value is apportioned to the previous period, and the
remaining 10s worth is held over for the next period. See
http://www.vandenbogaerdt.nl/rrdtool/process.php for a more detailed
Then in the first suggestions above, refer to the relevant bit, eg :
> "GAUGE does not save the rate of change. It saves the actual value
> itself. There are no divisions or calculations. Note though, GAUGE,
> like all other types IS normalized. See the section on "Normalisation"
* Is that correct ?
I have to admit that I found it hard to get to grips with rrdtool,
and these points are not too clear from the official documentation.
It might help if Ales's pages were linked from the official
documentation as well.
More information about the rrd-users