[rrd-users] Re: Clearing up some grey areas

[Dave Plonka]

Mark,

On Wed, Jun 21, 2000 at 10:58:13AM +0800, Mark Parker wrote:
> At present I am using a third party bandwidth software which I have
> configured to report out the 'ABSOLUTE' value (that is not the cumulative
> bps, but the bps for the last 5 minutes) of bps for different protocols etc,
> and I have RRDTool graphing this well. 
> 
> The difficulty I am having is that the counter is reset every night, and
> this appears to break my graphs, as after the nightly reset the graphs only
> show a single 'spike' of data at the time of the reset, and no other data
> seems to affect the outcome of the graphs.
<snip>
 
On Wed, Jun 21, 2000 at 02:33:37PM +0800, Mark Parker wrote:
> In my previous message I do not think I was clear about what the graph was
> displaying. All that the graph shows is a short 5 minute spike at 12:00 when
> the counter is reset ,and the rest of the graph both before and after that
> time is just simply blank.

It's probably just that the values are very low by comparison and seem
non-existent in the graph...

> Is there a way in the graph command to tell it to automatically graph from
> 12:05am to the present time on each day? this would elimate the period in
> the data source where the counter is reset.

While I don't think that's really what you want to do (as you could
never see data older than a day), you can just specify the start time
using "rrdtool graph --start".

However that seems rather simplistic... what you really need to do is
to avoid putting bad data into the RRD in the first place because,
whether you like it or not, RRDTOOL is going to use those bad values
(from ~midnight or whenever) when it consolidates the values and
percolates them into other RRAs containing hourly, daily, etc.
averages.  If those bad values are showing as spikes in the 1st RRA,
they're going to cause your averages to be exaggerated in subsequent
RRAs and graphs over longer periods of time.

If I'm understanding your first post at all, one way of addressing your
problem would be simply to set the `heartbeat' value to be very low
(like say 7 minutes) when you re-create the ".rrd" files.  (No point in
trying to tune them or dump and restore them to update the heartbeat
since it sounds like you've put some bad values in already).  Then be
sure to *skip* the update using the bad value you get at ~midnight.
This will cause RRDTOOL to record an *UNKNOWN* value there, which
sounds like the Right Think(tm) since you're saying that your
commercial tool is giving you a spurious value.

Dave

-- 
plonka at doit.wisc.edu  http://net.doit.wisc.edu/~plonka  ARS:N9HZF  Madison, WI