[rrd-users] Doubt in RRDDef

[Simon Hobson]

mini priyanka wrote:

>The method setStep is present in RRDDef.Can anyone let me know what is it?
>
>Say for example, if i set the step to default value 300s and 
>heartbeat for each DS as 20 minutes. And I get data only once in 15 
>minutes.In that case can I update the RRD every 15 minutes or should 
>I metion the step as 15 mins? Can I configure the step after the RRD 
>file is created?

I think you need to set the step when you create the database.

Before suggesting whether your settings may or may not be 
appropriate, you need to understand that they are.

STEP is the fundamental time period used by RRD Tools. Whatever value 
you use, RRD stores data in 'buckets' of exactly that length.

Heartbeat is a measure of how old your data can be before it's 
considered invalid. For some applications it may be valid to collect 
data intermittently, and use that to fill in a large period of time 
in the past. For example, you could have an electricity meter that 
just outputs a pulse every time a certain amount of electricity has 
been used - you count those pulses and update an RRD file each time 
you get one. When you're using very little electricity, the pulses 
may be a long time apart - but it's valid to say that over the 
previous 10 hours you averaged 1/10th of a pulse per hour. When using 
more electricity, the updates will be more frequent and the usage 
more detailed. Thus you may want a long heartbeat to allow this "old" 
data to be used.
On the other hand, say you were measuring the speed of a car. The 
speed measured at any instant is not really indicative of what speed 
you were doing a minute ago, and completely unrepresentative of the 
speed you were doing an hour ago. Thus a short heartbeat would be 
more appropriate.
BTW - these are not suggested as real situations - both could be 
handled better in different ways.

Lastly, you will need to consolidate the data. Say you were using a 5 
minute STEP size (300 seconds). That will allow you to plot graphs of 
data with 5 minute resolution. But over a year the quantity of data 
would be large, and the processing to draw a graph significant. For 
example, if you plotted a graph with one day/pixel then RRD would 
need to consolidate 288 samples for each pixel - every time you draw 
the graph. So RRD allows you to consolidate the data as it goes. If 
you ask it to consolidate 288 steps into one consolidated data point 
(CDP), then graphing is just a matter of mapping one CDP to one 
pixel. Not only that, but you could store a couple of years with only 
730 CDPs, rather than having to keep over 210,000 unconsolidated 
samples.


Alex has some excellent tutorials at http://www.vandenbogaerdt.nl/rrdtool/
Note especially the one on Normalisation and Consolidation.


Now, back you your specific question. If you have a step of 300 (5 
minutes), a heartbeat of 20 minutes, and update every 15 minutes ...
Then for each update, you will "fill in" the data for the previous 3 steps.

Whether this is right depends on what you want to do with the data. 
If you plot it then you'll find you have flat segments 3 steps long 
in the graph. If you never want data more detailed that 15 minutes 
then you may be better using a step of 900 (15 minutes).

On the other hand you can collect and update data as often as you 
wish. So you could collect data every few seconds, and RRD will 
simply accumulate it until a step is complete - at which point it 
will fill in the value for the step period just completed and start 
accumulating for the next. RRD will never store data more detailed 
than your step size - so even if you collected every second, you'd 
still only get one value per step.

-- 
Simon Hobson

Visit http://www.magpiesnestpublishing.co.uk/ for books by acclaimed
author Gladys Hobson. Novels - poetry - short stories - ideal as
Christmas stocking fillers. Some available as e-books.