[rrd-users] RRD PERCENT question (95 percentile)

Alex van den Bogaerdt alex at vandenbogaerdt.nl
Tue May 18 00:24:43 CEST 2010

> Ah that's what the 'real' issue was!
> I didn't realize that 'rrdtool graph' started to consolidate data based on 
> the width.

else it would have to display two or more rates in one pixel column...  how 

> So if I understand correctly, it uses 1px for 1 sample? if there are more 
> samples then pixels it will consolidate the data?
> Is there a way to let RRD tool figure the width out automatically?

not that I'm aware of, but things may have changed since I last hacked code.

> without consolidating the data? Reading the docs it seems --full-size-mode 
> could help me here?
> I couldn't test that as the version I run (1.2.19) does not recognize that 
> option.

According to the docs, that option specifies the width of the entire image 
created. You end up with a graph size which fits inside it.  So no, it would 
actually make your job harder. Instead of having 400x100 [x] pixels inner 
graph area and a surrounding part, you have a 400x100 [x] image with some 
unknown inner part and thus an unknown amount of columns.
([x] for 400x100 you can read anything specified with width and height).

It's not hard. You know (or should know) which RRA you use, thus what the 
duration of each CDP is, so you can do as I did:
intervals = (endtime-starttime)/duration_of_CDP

I prefer doing so anyway, for all graphs. And this also forces you to think 
about your RRA strategy when creating your RRD.


More information about the rrd-users mailing list