[rrd-users] Newby questions

[Pablo Chacin]

Hi Tobi

> 3. Finally, is there any patch to add SUM function for totalizing a
> metric over a period. This is a very important requirement as we need to
> count the total number of certain events for minute, hour and day. I'm
> aware of the approach of using the average and multiplying by the
> reporting period, but this will lead to a approximated value.
>    everything in rrdtool is a 'rate' so even if there was such a
>    function the math would still be the same ... can yo elaborate in
>    what way you expect the result of the calculation to be an
>    aproximation?
>

According to what I read in the previous discussions on this list
(https://lists.oetiker.ch/pipermail/rrd-developers/2005-April/001405.html)
The presence of  unknown values may produce wrong totals when
calculating the total from the average. For example:

>    0:UNKN 300:10 600:10 900:10
>
> consolidated and then mulitplied by the period will produce the same value
> as if all entries were 10, and not with one UNKN (given the cf is 0.5). A
> SUM function would nicely return 30.
>

Regards

Pablo



On 10/29/2012 10:16 PM, Tobias Oetiker wrote:
> Today Pablo Chacin wrote:
>
>> Greetings all
>>
>> I'm new to rrdtools. I'm actually still evaluating it as as tool for
>> storing monitoring metrics.
>> After reading the documentation and several discussions in this list and
>> also in other forums
>> I'm still confused about a couple of things.
>> 1. How are subsecond timestamt reported? In several places it is
>> mentioned the existence of these timestamps, but can't find any example
>> of how to specify it in the update command, except when using "N" as
>> timestamp.
>    rrdtool update 1351545246.3897:394.33
>
>> 2. Supposing subsecond timestamps are supported by rrdtools, are they
>> also supported by rrdcached?
>    I think the version on github does support it ... the released
>    version not.
>> 3. Finally, is there any patch to add SUM function for totalizing a
>> metric over a period. This is a very important requirement as we need to
>> count the total number of certain events for minute, hour and day. I'm
>> aware of the approach of using the average and multiplying by the
>> reporting period, but this will lead to a approximated value.
>    everything in rrdtool is a 'rate' so even if there was such a
>    function the math would still be the same ... can yo elaborate in
>    what way you expect the result of the calculation to be an
>    aproximation?
>
>> Even when
>> I understand the concerns about it not working properly in all
>> situations, I think is a very common requirement. Simply can't believe
>> nobody has done this ever, unless I'm completely wrong about the need of
>> such function.
> cheers
> tobi
>
>> many thanks in advance
>>
>>
>>


-- 
Pablo Chacin
R&D Engineer
SenseFields SL
Tlf (+34) 93 418 05 85
Baixada de Gomis 1,
08023 Barcelona (Spain)
http://www.sensefields.com/