[rrd-users] consolidation or no consolidation?

[Simon Hobson]

KOZMAN Balint wrote:

>Well, on the presentation side (regarding rrdgraph) I can ensure, that
>number of values to be presented will fit. To be more precise, I have a
>worker thread somewhere which inserts recently collected data to the rrd
>on a 10 sec period, having only 1 DS in the rrd. Now, when drawing a graph
>of the last 1 minute of this DS, all I have is exactly 6 entries in the
>rrd. Thus I'd like to see those values on the graph, and not something
>that just compares to them somehow. Obviously when graphing a month-long
>period, I would like to see consolidated data as that's the only thing I
>can expect.
>So what parameters should I use during rrd creation to get this? A small
>example would help a lot.


You need an rrd with a step size of 10 AND you must feed it data 
points with timestamps exactly divisible by 10 - about 10 s is not 
good enough, it has to be EXACTLY 10s. You need a consolidation 
function in the rrd with ONE step per consolidated step - is matters 
little which you use as min, avg, and max are the same for one single 
value.. If you then graph that rra with start and end times which are 
themselves exactly divisible by 10s AND number of pixels exactly 
equal to number of samples AND end time which is no later than last 
complete step time AND the start time is within the range stored by 
the hi-res rra THEN you should get exactly the same values out as you 
put in.

What is happening if you meet all these requirements is that you will 
avoid normalisation (input data already matches database time slots) 
and avoid consolidation (or more technically perform null 
consolidation) by having a CF that is only one sample long.