[rrd-developers] Re: Restarting devices

[Alex van den Bogaerdt]

> >
> > - at time 949001506 the counters are unknown
> > - at time 949001507 the counters are known to be zero
> > - at time 949002007 the counters are known to be *very* large again
> >
> > In 500 seconds, the counters increase by 2848208129 and by 2744819897.
> 
> Do you mean 2848207648 and 2744819601, which are the counter values for the
> update that occurred at time 949002007?  (It doesn't make much difference to
> the arithmetic, either way, but I want to make sure I understand what is
> happening.)
> 

It does make a difference normally.  The numbers are not just the counter
values, they are the differences.  As you start out with zero, the delta
(the difference) is the same as the current value.  This won't happen during
normal day to day operation.

How does your average program (including the front end to rrdtool) compute
the rate at which data flows over your lines?
It takes a sample S1 at time T1, it takes another sample S2 at time T2.
The amount of data that was passed is (S2-S1), it takes (T2-T1) seconds 
to do this.  For example: (2848208129-0) bytes in (949002007-949001507)
seconds.  This means that *on average* the amount of bytes per second is
5696416.

For internal computing, rrdtool assumes that each second this amount of
data passes the device.  Your sample was taken on time 949002007.  RRDtool
normalizes the data so that it can be stored on 5-minute intervals.
If you divide 949002007 by 300 (300 seconds is 5 minutes) and use whole
numbers only, you get 3163340.  Multiply this with 300 to get 949002000.
This is the time at which RRDtool stores the normalized sample.  Your
sample was taken 7 seconds after this time; these 7 seconds were calculated
to carry 5696416 bytes each.  7*5696416=39874912 bytes in 7 seconds.
Assuming the rest of the interval no data is transfered at all, these
39874912 bytes are transfered in 300 seconds.  After doing the math, you
get an average of 39874912/300=132916 bytes per second.

> The question I initially meant to ask is still dangling:
> The average values 5696416 and 5489639 obtained for the 500-second interval
> just following the presumed counter reset (or even their adjusted 300-second
> interval values, I believe) exceed the maximum value (256000) I specified for
> the DS;  why does RRDtool go through with the calculations for the next
> interval, using these 'bad' values?  Shouldn't it come up with NaN for the
> 949002300 interval?
> 

One could argue that this rate of 5696416 per second is absurd and that
this should therefore be ignored.  Obviously RRDtool doesn't do this, or
my understanding of the tool is insufficient.

Tobi, if you're reading this line here in this mail, I apologize for the
redundant mail to come next :-)

Regards,
-- 
   __________________________________________________________________
 / alex at slot.hollandcasino.nl                  alex at ergens.op.het.net \
| work                                                         private |
| My employer is capable of speaking therefore I speak only for myself |
+----------------------------------------------------------------------+

--
Unsubscribe mailto:rrd-developers-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-developers-request at list.ee.ethz.ch?subject=help
Archive     http://www.ee.ethz.ch/~slist/rrd-developers