[rrd-users] Re: data resampling Q

Tobias Oetiker oetiker at ee.ethz.ch
Thu Oct 14 21:30:12 MEST 1999

Today you sent me mail regarding Re: [rrd-users] Re: data resampling Q:

*> On Wed, 13 Oct 1999, Tobias Oetiker wrote:
*> > *> so the closer i get `rrdtool update` to the rrdtool 300 sec interval
*> > *> (epoch time), the more accurate my samples will become?
*> > 
*> > sure ... but if you need more acuracy you should decreass the sampling
*> > interval so that sampled data shows what you are looking for with sufficient
*> > accuracy ... aiming for the 300 points is not the right aproach here ...
*> true.  i guess i meant to say precision, not accuracy.  i cannot change
*> the sampling rate. 
*> -snip-
*> I'd rather know that "at approximately 14:45 there were exactly 20 users
*> logged on", than "at exactly 14:45 there were about 20.534 users logged
*> on".  This is particularly confusing when it is only possible for a
*> maximum of 20 users to log on at one time! 
*> I've even considered adding a new, un-interpolated type (an ABSOLUTEGAUGE
*> perhaps??) to rrd.  What do you think?
*> -snip-
*> i'd love it.  and if more than one sample comes in the interval just
*> have it keep the most recent . . 

no this is totally against the spirit of things ... it does not make sense
... and I would be most interested to see how you can get an average of
20.534 when you only input values of 20 or less ... because this would be a
bug ...

*> or maybe a mod to rrdtool update that would would round to the nearest
*> interval.
*> e.g.
*> rrdtool update load.rrd NINT:111

I suggest you do this in the script supplying the data ... you can replace N
with a unix seconds since epoch counter ... and you can use MOD to alter
your data ... but realy, it does not make any sense at all to artificially
add further jitter to your data ... if you aquire a sample at 3 seconds past
the hour you should tell rrdtool and rrdtool will do its best to determine
the value of the sample at the hour ...

one thing could be done in this area ... at the moment the GAUGE datasource
type assumes that a DS has been on the value supplied since the last update
... for some application you might want to interpolate the curve in between
samples as straight lines instead .. 


*> --
*> Scott Walters
*> -PacketPusher
*> "The world speaks IP"

 ______    __   _
/_  __/_  / /  (_) Oetiker, Timelord & SysMgr @ EE-Dept ETH-Zurich
 / // _ \/ _ \/ / TEL: +41(0)1-6325286  FAX:...1517  ICQ: 10419518 
/_/ \.__/_.__/_/ oetiker at ee.ethz.ch http://ee-staff.ethz.ch/~oetiker

* To unsubscribe from the rrd-users mailing list, send a message with the
  subject: unsubscribe to rrd-users-request at list.ee.ethz.ch

More information about the rrd-users mailing list