Spreadsheet for calculating RRA values

[portante]

Attached is a spreadsheet that I created while trying to wrap my mind around how RRAs in RRDs work. I am posting it here in the hopes that others might correct any errors in it, and/or correct my understanding, or be a useful resource for others.

The spreadsheet has three sheets in it:

Example: which just lists out the numbers from the RRD tutorial at: http://oss.oetiker.ch/rrdtool/tut/rrdtutorial.en.html, in such a way that I could understand the math and play around with the numbers

CactiRRAs_Default: which list our Cacti default RRAs (which are most likely not the original defaults, but were broken, and the catalyst for me to work to understand how all this stuff fits together)

CactiRRAs: presents a way to calculate a set of five RRAs for viewing data

Here are a few notes about the CactiRRAs* sheets:

CactiRRAs_Default: I noticed that the views of the graphs in our system seemed odd for some reason. So I decided to calculate out the timespans for the collections. It turns out the timespans in the RRAs don't match the representation of time actually collected in the rows. This sheet shows the correction, and the difference. Also, we had tried to change our system to 60 second samples, from 5 min (300 sec) samples, but did not change these RRAs. So with this sheet if you change 300 to 60, you'll see how silly the RRAs look. It seems odd that Cacti stores the time between samples in the data template, but the RRAs rely on that sample size to get expected results. Perhaps an option to have timespan automatically calculated, or define sets of RRAs that work with a given sample size (one set of 60 sec samples, another set for 300 sec samples). Or maybe there is a way to do this that I don't understand. Please correct my understanding here.

CactiRRAs: The default numbers here keep the size of the RRDs down, yet given us higher resolution data over a 2 year period. We wanted 1 month of 1 min samples, and then additional time periods to be much longer so that the resolution tailed off linearly. We felt that 10 min samples for 205 days was good to still be able to make good historical comparisons. Also note, that the sheets let you pick whether or not the RRAs include AVERAGE, MIN, MAX or LAST in them as well, so that you'll see the total rows of data collected to help get an understanding of how the size of things is impacted. For the 1 min samples, since we have a step of 1, there is no point storing MIN, MAX or LAST, so we just store AVERAGE there. But we add MIN and MAX for the other RRAs so that we can use it in graphs for historical analysis.

Hope this is useful to folks, and please don't hesitate to clarify anything that I did not get right here.

Thanks, -peter