Difference between the values read and the values stored

david.knecht · Post by **david.knecht** » Mon Sep 22, 2003 3:23 am

Hi all,

I'd like Cacti 0.8.3 to graph ***exactly*** the value that it got as its input.

I am using a GAUGE data type and the LAST consolidation function.

This is what the Cacti log is showing (thus, these are the values I'd like to see in the graphs):

09/22/2003 9:40 AM - CMD: /opt/rrdtool/bin/rrdtool update /var/www/html/cacti/rra/RTTSum-rtr_1.rrd --template RTTSum N:4726
09/22/2003 9:45 AM - CMD: /opt/rrdtool/bin/rrdtool update /var/www/html/cacti/rra/RTTSum-rtr_1.rrd --template RTTSum N:4102
09/22/2003 9:50 AM - CMD: /opt/rrdtool/bin/rrdtool update /var/www/html/cacti/rra/RTTSum-rtr_1.rrd --template RTTSum N:4891
09/22/2003 9:55 AM - CMD: /opt/rrdtool/bin/rrdtool update /var/www/html/cacti/rra/RTTSum-rtr_1.rrd --template RTTSum N:3855
09/22/2003 10:00 AM - CMD: /opt/rrdtool/bin/rrdtool update /var/www/html/cacti/rra/RTTSum-rtr_1.rrd --template RTTSum N:4724

And these are the values that are stored in the round robin database (output of the "rrdtool fetch" command):

1064216400: 4.7134000000e+03
1064216700: 4.1144800000e+03
1064217000: 4.8725900000e+03
1064217300: 3.8722666667e+03
1064217600: 4.7066200000e+03
1064217900: nan

As you can see there is a slight difference between what Cacti reports and what rrdtool stores/graphes (in fact, each value differs towards its preceding value).

Any ideas what can cause this difference?

Thanks, David

Post by **raX** » Mon Sep 22, 2003 10:51 pm

The slight difference in values that you are seeing is due to RRDTool. By default the step for each data source is set to 300 seconds, which means RRDTool expects new data every 300 seconds. In reality this ends up being 299 or 301; something really close to 300 seconds. RRDTool compensates for this small disparity by adjusting the number it receives as input.

-Ian

david.knecht · Post by **david.knecht** » Tue Sep 23, 2003 5:59 am

Ian,

This is what I was thinking, too. However, I assume that a value of type GAUGE is not related to time, thus, there is - as to my current understanding - no need for such a compensation. Secondly, given that data is acquired every 299/300/301 etc. seconds, why would then *every* value tend towards its preceding value? It seems to me that there is some averaging mechanism inside RRDTool that I have not understood so far so I will raise this issue in the RRDTool forum.

Ian, thanks for your help!

David

david.knecht · Post by **david.knecht** » Wed Sep 24, 2003 2:24 am

For the sake of completeness - it might help somebody, too...:

(copied from the RRDTool forum at http://www.ee.ethz.ch/~slist/rrd-users/msg07347.html)

On Tue, Sep 23, 2003 at 01:32:51PM +0200, David Knecht wrote:

> I'd like Cacti 0.8.3 and RRDtool 1.0.39 to store and graph ***exactly***
> the value that it gets as its input.

Then you have to tell RRDtool to store at ***exactly*** the right time.

> As you can see there is a slight difference between what Cacti reports
> and what rrdtool stores/graphs (in fact, each value differs towards its
> preceding value).

Normalization.

> Does this mean that RRDTool does not/cannot store values in its raw
> form, even if using GAUGE data types? Is it correct that RRDTool is
> doing some time compensation even for GAUGE data types (which to my
> latest understanding are not related to time?) Might any other averaging
> mechanism fool me here?

COUNTER -> rate = delta(counter)/delta(time)
ABSOLUTE -> rate = (value)/delta(time)
GAUGE -> rate = value.

Then:

rate -> normalization -> RRA.

*If cacti is using something similar to "rrdtool update N:12345" then
what is "N"?

N is approximately "09/22/2003 9:40 AM". N is not exactly "09/22/2003 9:40 AM"

HTH
Alex