[rrd-users] Re: disabling interpolation?

Alex van den Bogaerdt alex at slot.hollandcasino.nl
Wed Aug 23 01:06:34 MEST 2000

Philip Molter wrote:
> On Tue, Aug 22, 2000 at 06:46:29PM +0200, Alex van den Bogaerdt wrote:
> : Right.  As I said: don't mess with the averages.  However, it is safe
> : to say the minimum amount of people in that room is 4, not 4.76 and
> : the maximum amount of people in that room was 5, not 4.76 .
> : It makes no sense to say that during this interval there were 4.76
> : people in the room *unless* you are talking about averages.
> Wrong.  You can't say that at all.  You could very easily say that
> the minimum amount of people in the room as 0 and the maxium amount
> was 12, and I could still give you a scenario where the average as
> 4.76

Ack, this can happen.  Also, you never know for sure that it did or did
not happen (with our way of monitoring).  At least I *can* be sure that
the *actual* number of people (and *not the average*) was not 4.76  This
is what the discussion is about:  is it proper to first average and then
use max/min on it?

Averaging the value first also introduces an error.  Assume the following
number of people:

12:00  12 (for the sake of this example this number doesn't change until...)
12:05  12
12:06   1
12:07  10
12:08  34
12:09   9
12:10  10
12:11  15
12:12  14

Samples are taken at times 12:02, 12:07 and 12:12.
Calculated averages (not verified, doesn't matter for this discussion):
12:00..12:05    12*2/5 +  1*3/5 =  5.4
12:05..12:10    10*2/5 + 14*3/5 = 12.4

If the proposed data type would be available:
12:00..12:05    12
12:05..12:10    10

Explain to me why 5.4 is better than 12 ?
Then again, min(12:05..12:10) should be 1 and the proposed DST cannot
do this either. 

Increasing the frequency of your polling will increase the chance that
you happen to sample at the correct time.  It does *not* improve
accuracy.  Depending on the time when I run the script I get either
10 (@12:07) or 34(@12:08) as input.  Granted, I get both values if I
sample every minute.  No problem ...
   12:07:00  value 10
   12:07:01  value 99
   12:07:02  value  0
... and then this argument is not valid.

Clearly, in the concept of RRDtool the assumption is made that gauges
do not differ this much (or perhaps: the user should make this assumption
or do not use RRDtool).  If the data does not change a lot, the original
example numbers (4..5) are more likely than yours (0..12).  If it is okay
to make this assumption then:

The example used numbers 4 and 5.  It may or may not be correct to assume
4 and 5 were the minimum and maximum.  I agree on this.  However, if I
sample those two values, and use min/max functions on them, I would at
least keep them to 4 or 5.  I never got 4.76 and therefore I do not want
to see 4.76 ...

As discussed elsewhere on the list: *I* know how to make this happen.
It's just that I think a computer program should be doing the job, not
a human.

 / 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 |
| http://faq.mrtg.org/                                                 |
| http://rrdtool.eu.org  --> tutorial                                  |

Unsubscribe mailto:rrd-users-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-users-request at list.ee.ethz.ch?subject=help
Archive     http://www.ee.ethz.ch/~slist/rrd-users
WebAdmin    http://www.ee.ethz.ch/~slist/lsg2.cgi

More information about the rrd-users mailing list