[rrd-users] Simple DEF Arithmetic

Chris Mason chris at netnix.org
Wed Aug 7 10:57:15 CEST 2013

Hi Johan,

Even if data is consolidated - you will get the max-values - as your RRD's
> are set up to keep these.
> But looking on a wider scope - you may get a difference on the sum of the
> Max - I think... [image: :-)]
> Another example. of single samples.
> A:  1 2 3 3 2 1
> B:  3 2 1 1 2 3
> For each of the above - the sum is always 4 - when adding up A and B
> numbers.
> But when all the 6 values are consolidated to a single value - and you use
> MAX - you will end up having 3 for A - and 3 for B.
> The sum of these are 6 - (and not 4).
> So if larger buckets are created - and the sum of these are calculated you
> will end up with a higher number.

Thank you, that is an easy to understand explanation that seems to
demonstrate what I am experiencing.
I couldn't get my head around it until I saw that.

Basically, my 1 month graph uses a 2 hour consolidation period and my 1
year graph uses a 1 day consolidation period. It is unlikely that MAX(A)
and MAX(B) are going to happen at the same time. Let's assume, on my
monthly graph that MAX(A) happens in a different 2 hour bucket to MAX(B),
but on my yearly graph, MAX(A) and MAX(B) occur within the same bucket as
the buckets are larger - the MAX on the yearly graph will be higher than
the MAX on the monthly graph. Does that sound plausible?

I think I have just convinced myself of the pit falls of using MAX when
combining data sources.

For network bandwidth it's normal to look at the 95%-percentile

Thanks, I will have a look into this.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.oetiker.ch/pipermail/rrd-users/attachments/20130807/2445ea73/attachment.htm 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/png
Size: 1057 bytes
Desc: not available
Url : http://lists.oetiker.ch/pipermail/rrd-users/attachments/20130807/2445ea73/attachment.png 

More information about the rrd-users mailing list