[rrd-users] cant get to understand described normalization behaviour

Alex van den Bogaerdt alex at vandenbogaerdt.nl
Wed Mar 26 10:47:16 CET 2014

I think you're missing the fact that RRDtool uses timestamps based on UTC.
When you specify "--step 86400" you are telling RRDtool to create a database 
with start and end times which are a whole multiple of 86400 seconds.

Your first timestamp is 1395140400 which is n*86400+39600.
Your second timestamp is 1395226800 which also is n*86400+39600.
I did not check the rest but I can safely assume the rest will not be 

Normalization massages the input so that it fits in (n-1)*86400 to n*86400 
time intervals.

The rate computed at 1395226800 is valid partially in the interval upto 
1395187200 and partially thereafter. To be precise, (86400-39600)/86400 
stays in the timeslot computed for 1395187200 and 39600/86400 goes to the 
next timeslot.

Your next update is at time 1395313200. Now there is enough data to compute 
the timeslot between 1395187200 and 1395273600. The rate logged at time 
139573600 should be 39600/86400 * 3 + 46800/86400 * 2 = 2.458333....


----- Original Message ----- 
From: "Peter Thurner" <p.thurner at heinlein-support.de>
To: <rrd-users at lists.oetiker.ch>
Sent: Wednesday, March 26, 2014 10:24 AM
Subject: [rrd-users] cant get to understand described normalization 

Hi guys,

I'm kind of dying on this one. I attached 3 bash scripts that show where
I don't understand things anymore. I guess its quite simple.

I feed the same values using the same CF and RRD specs everywhere, but
on the script rrdtest-days.sh, I get a graph that is a lot different to
the other two scripts.
Other than the --step size I did not change much (as a diff would show).

Why (and how?) is it normalizing when all I increase is the timespan
between updates? I'm sure I missed something..

Thank you very much in advance! :)



> _______________________________________________
> rrd-users mailing list
> rrd-users at lists.oetiker.ch
> https://lists.oetiker.ch/cgi-bin/listinfo/rrd-users

More information about the rrd-users mailing list