[rrd-users] A reason there isn't a CF SUM function?

[Tobias Oetiker]

Hi Derek,

Aug 22 Derek Haynes wrote:

> Tobias,
>
> > in any event, on a technical level, what would be required to
> > achieve what you want (appart for the handling of unknown periodes
> > within a single step) would be an AVERAGE_UNKNOWN_AS_ZERO
> > consolidation function ... this would not break the assumptions
> > about the nature of RRAs and you could calculate your type of SUM
> > easily ...
>
> Makes sense.
>
> >   a + unknown = a
>
> This seems to have been addressed in other areas (ADDNAN / TOTAL
> w/VDEF) both appear to treat unknowns as zero. An AVERAGENAN operator
> for a CDEF would seem to complete the loop.
>
> Any suggested starting points for adding this to rrdtool? I haven't
> dug through the source before.

search for the name of the consolidation function in the source ...
it is sort of orderly ...

cheers
tobi

>
> Cheers,
>
> Derek
>
> On Wed, Aug 22, 2012 at 2:50 AM, Tobias Oetiker <tobi at oetiker.ch> wrote:
> > Hi Derek,
> >
> > Yesterday Derek Haynes wrote:
> >
> >> Several times on the list, the topic of a CF SUM function has
> >> risen...just a sample:
> >>
> >> http://oss.oetiker.ch/rrdtool/forum.en.html#nabble-td1059458
> >> http://oss.oetiker.ch/rrdtool/forum.en.html#nabble-td1059458
> >> http://oss.oetiker.ch/rrdtool/forum.en.html#nabble-td1059458
> >>
> >> I'm in need of a similar CF SUM function and I was curious if there
> >> isn't one because (1) there's a fundemental reason why it's not
> >> provided or (2) it just hasn't been implemented.
> >
> > the fundamental problem with a SUM CF is that the values of the RRA
> > entries would then depend on the step interval of the RRA and thus
> > fundamentally change the nature of the RRAs ...
> >
> >> For reason (1), the common argument is you can calculate this by
> >> multiplying the average by the duration of the CDP. However, this
> >> doesn't hold when a CDP contains unknowns as the average is only based
> >> on the known values.
> >
> > basing a sum on unknown values might be problematic too
> >
> >   a + unknown = a
> >
> > would mean that you assume that unknown equals 0
> >
> > where we curently make the assumtion that the unknown part of the
> > periode was similar to the known parts ...
> >
> > in any event, on a technical level, what would be required to
> > achieve what you want (appart for the handling of unknown periodes
> > within a single step) would be an AVERAGE_UNKNOWN_AS_ZERO
> > consolidation function ... this would not break the assumptions
> > about the nature of RRAs and you could calculate your type of SUM
> > easily ...
> >
> > cheers
> > tobi
> >
> >
> >> The TOTAL operator does exactly what I'm looking for but it's only for
> >> VDEF statements.
> >
> >
> >> Any thoughts?
> >>
> >> Derek
> >>
> >>
> >
> > --
> > 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
>
>
>
>

-- 
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