[rrd-users] updates, integers, double, polating, interpolating, whatever: FAQ

Alex van den Bogaerdt alex at vandenbogaerdt.nl
Wed Nov 12 11:13:13 CET 2008

To all new users:

rrdtool will force your data to fit in pre-defined slots.  It's called 
normalization and is explained all over the net, including here: 

You may want to try a step size of 1 (1 second) combined with an RRA which 
stores 300 of such steps per entry (RRA:AVERAGE:0:300:420768 or similar).

You may also try changing "rrdtool update filename.rrd N:123:123" into 
"rrdtool update filename.rrd 1226484600:123:123" where 1226484600 is a 
timestamp in seconds since the epoch, see for instance the unix date 
program. That timestamp should be an integer multiple of your step size 
(that's why step==1 should work for any timestamp, were it not for subsecond 
precision, not sure if it still works or not).

And then there's consolidation, also explained there, which may further 
'modify' (not really!) your data.

That is because you are working with rates.  If 12 men visit your store in 
one day, you have _on_average_ 0.5 men _per_hour_ walking through that door. 
Yes, 0.5 is perfectly valid, eventhough you will never actually see half a 
person on the street.  (you may encounter short people, which is something 
entirely different!)

If those 12 men enter between 08:00 and 08:30, then during that half our you 
have an average rate of 24 men per hour. If during the rest of the day 
nobody enters, you still end up with 12 men in one day, which is 0.5 men per 
hour. That's why *averages* can go down from 24 to 0.5 ...

If you are looking at MAX instead of AVERAGE, then 24 will stay 24.


More information about the rrd-users mailing list