[rrd-users] Trend of Daily Peaks

[Karl Fischer]

Am 09.02.2011 21:00, schrieb Kyle Brandt:
> If possible I would really like to do it all in the graph.
> 
> To define the peak mathematically something like the following would
> be ideal. Break the day into hour blocks and average out of the
> bandwidth usage of the blocks. Then take that result and pick the
> highest value block as the peak value for the day.  Lastly, trend
> these values over a larger period of time. This seems like something
> that would be easy to do in the rra originally -- but I want to work
> with existing data.
> 
> Does that make sense?

well, somewhat ...

You can break the data timeline into hour blocks by specifying the step,
optionally followed by a consolidation function for data reduction.
So in theory you could be using something like this:

DEF:ds0=sample.rrd:ds0:AVERAGE
DEF:ds0hourly=sample.rrd:ds0:AVERAGE:step=3600:reduce=MAX

you can pick the daily max the exact same way:

DEF:ds0daily=sample.rrd:ds0:AVERAGE:step=86400:reduce=MAX


however, I'm not quite sure if reduce=MAX would work as expected in
combination with the step=3600 or if these values would be averaged?
I guess it's worth a try.

Of course you're better off if you already have the MAX values in rrd.
Then you can use:

DEF:ds0=sample.rrd:ds0:AVERAGE
DEF:ds0hourly=sample.rrd:ds0:MAX:step=3600
DEF:ds0daily=sample.rrd:ds0:MAX:step=86400

Then you can create your trend on the hourly MAXes:

CDEF:ds0trend=ds0hourly,86400,TREND

give it a try and see where you get ...

- Karl