[rrd-users] Multiple RRA for pdp_per_row of 1

[Larry Low]

> > > I knew there was something I was missing.  If you try to graph MAX
> and
> > > only have AVERAGE rra for pdp_per_row it will use the first MAX rra
> it
> > > finds instead.
> > >
> > > Is there a way around this without forcing the resolution or do I
> need
> > > to store the MAX rra and just fill it with the AVERAGE data?
> >
> > Any ideas?   I am guessing I'm going to be forced to duplicate my
> AVERAGE
> > rra into a MAX rra for pdp_per_row 1.
>
> Since, at pdp_per_row=1, MAX is effectively the same as AVG, your MAX
> RRA
> would be redundant except that it would, as you say, prevent a higher
> pdp_per_row MAX RRA from being taken in preference if you ask for a MAX
> at
> this resolution.  However, why would you be asking for a MAX at this
> resolution?  If you're coding this yourself then just take the AVG
> instead
> as it is the same data - this is why MRTG doesn't display peak lines
> for
> 'Daily' graphs.
>
> On reflection, though, I notice that MRTG DOES create a pdp_per_row=1
> MAX
> RRA by default as I think it helps with other time-period MAX
> calculations -
> if you ask for a single value of max(ds) between t0 and t1, I believe
> it
> looks for a MAX RRA with the least overlap from (t0,t1] to calculate
> from.
> This may be a candidate for RRDtool optimisation (if you need a
> pdp_per_row=1 MAX RRA and one is not available, then a pdp_per_row=1
> AVG or
> LAST will do just as well)
>

Yes this is exactly what I was thinking.  The big benefit of rrdtool is that you shouldn't need to know what data is inside the rrd file and it should use the best available data.  I usually only have 3 or 4 pdp's defined so removing the need to redefine a MAX,MIN,etc for pdp_per_row of 1 reduces file size by at least 25%.