[rrd-users] Bits/sec and GB/month
Tobias Oetiker
tobi at oetiker.ch
Mon Oct 20 17:38:38 CEST 2008
Today Simon Hobson wrote:
> Eric wrote:
>
> >As far as I understand, RRDTool never saves real data but only data from
> >points in time. So, if somebody is making graphs with 'Bits/seconds' and
> >stores this data in an rrdtool database, you cannot calculate back to totals,
> >unless assuming the average for the default time of 5 minutes, was the same
> >those 5 minutes and multiply this average by 5*60. By this, bandwidth bursts
> >are heavily punished.
>
> You misunderstand. The input data is normalised and (if neccessary)
> converted to a rate. If you feed in a counter from an interface (eg
> octets in/out) then it will convert these values to a rate. Although
> the rate stored will be 'slightly out' if you don't feed it numbers
> exactly on a step boundary, over a period of time it WILL work out
> correct.
>
> Eg, suppose you have a step period of 300s (ie 5 minutes), and your
> rate is 300bytes/sec. The stored rate will be 1.
no ... it would store 300 since the stored rate is always PER
second. So for your example it should be 1byte/s.
> Lets suppose that
> for 10 minutes you double your rate, but your updates are out of sync
> with the rrd database. For the 10 mins, your rate is now 2, but it
> would be stored as 1.5 for 5mins, 2 for 5 mins, and 1.5 for 5mins.
> Over an individual period you have errors, but over a longer term,
> the average is still correct.
>
> Over a period of days/months it will work out correct and you can
> simply multiply average rate * time to get total - and apply a factor
> of 8 if required to convert between bits and bytes.
>
> >So, since they are telling me their totals are coming from their rrdtool
> >database, I think they provide me impossible values.
>
> As long as the data going in is accurate, then the totals coming out
> should be correct.
>
>
--
Tobi Oetiker, OETIKER+PARTNER AG, Aarweg 15 CH-4600 Olten, Switzerland
http://it.oetiker.ch tobi at oetiker.ch ++41 62 775 9902 / sb: -9900
More information about the rrd-users
mailing list