[rrd-users] Re: inserting historical data to rrdtool

[Craig Wyllie]

"BAARDA, Don" wrote:
> "Adam Treister" wrote:
>> If rrdtool preallocates the whole db when its created, seems like
>> it'd be simple to allow it to insert the data in any order.  When
>> the data is coming from disperse sources or overlaps between files,
>> this would make life easier.    Not a whole lot, but a little.

>         The problem is that rrd does adjustments to the data based on
> previous samples before accumulating and storing info into the database. It
> adjusts slightly miss-timed samples by calculated adjusted values using the
> previous sample (and it's time). It can fill in multiple entries from a
> single sample if the time-gap since the last sample spans multiple entries
> (and heartbeat is set high enough). An unknown sample results in unknown
> entries before and after that sample. etc. etc.
> 
>         For RRD to get this right, it needs to know _all_ the samples taken
> immediately before, immediately after, and during a recorded entry. If the
> samples are not time-ordered, how can it know when it has all the samples
> needed to fill in any entry?

In that case we'd have to force a discontinuity (in the form of a single
unknown) at the start and end of the imported data.

This does mean that when you import your data, you'd have to do it in
chunks that are big enough so as to not end up with a DB half full of
unknowns.

I wonder if there isn't some methodology or algorithm that would help sew
the ends together?  Not mathematically 'perfect', but 'best
effort'/most-likely...

To the first order, you could just ignore the discontinuity, go around
the averaging, and force the endpoint data into those tail slots.  (I
have *no* idea as to what that would entail nor it's plausability.. in
the code.  Flame away.)  That would probably be nicer (from the front
end/view POV) instead of putting in the unknowns at the tails of the
imported data.

Cheers

	-Craig

( Or one of us massochists implements 'actual-value' storage.  No, I'm
not goint to talk about it, I've never even looked into it. I just don't
have the time. Sorry :)  It does keep popping up. Someday someone will
come along *and* have the time. )

________________________________________________________________________

 FREE 30-day trial versions of modules for Sun Management Center 2.1.1

  Oracle     WebPortal  Linux   ScriptRunner  Support    Resellers
  Sybase     NT Agent   HP-UX   EventAction   Service    Ask an Expert
  Rep Srvr   MS-SQL     Cisco   Trap Fwding   Training   Custom Modules
________________________________________________________________________

     Craig Wyllie            |      mailto:craig at HalcyonInc.com
     Software Developer      |      phoneto:416-932-4606
     Halcyon Inc.            |      faxto:416-932-4711
     Toronto, Canada         |      http://www.HalcyonInc.com
________________________________________________________________________

--
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