MRTG Graph

[Michael Janssen]

On Sat, 5 Oct 2002, Matt wrote:

> > when i understand correctly, that the length (the shortness ;-) of the
> > intervall is your concern, i want to say, that i've done it with 5min
> > steps a few days ago.
> >
> > you can find a first sample here:
> > http://www.rz.uni-frankfurt.de/~janssenm/logstats/mail.gif
> > or mail2.gif for best resolution
> >
> > im still wondering if i can reproduce it a portable and flexible way or
> > how to set up rrdtool-graphics to give usefull impressions (time-period
> > covered, resolution, labels of axis, several graphs on 1 gif and so on).
> >
> > Tell me, when this is what your're looking for (otherwise i assume that
> > the kind of graphic people suspect from MRTG is something completly
> > different ;-).
>
> I have MRTG monitoring several t1's.  Green is downstream and a blue line is
> upstream.  That works and looks well.  What I would like to do on
> Mailscanner is have a green line for total emails, blue for spams and a red
> line for virii.  5 minute, 1 hour averages don't care.  I would just like
> something that looks like MRTG does monitoring my t1 circuits.  Would give
> me a visual indication of how things are working without poking around in
> the log files manually.  I already have MRTG installed on my Raq4i so I was
> hoping to use that.
>
> Matthew
>

5min against 1hour averages looks better, when you put the in an
MRTG/rrdtool-graph. But when you want to get an image covering only one
hour you need even more samples. With 5min resolution you will get 12
datapoints on 400 pixel. But when we reduce resolution to far we try to
get better than maillog does: the graph will randomly float.

The question is, what time-period should be represented by the serveral
images.

the only working mrtg-setup i ever seen tries to covers common periods
like day, week, month, year. This is resonable from a certain point of
view (however not my point of view, since i havn't got behind this whole
mrtg-story).

This means:
           resolution   samples    period
1. image      5min      300-400     25-33h
2. image     30min       400        8.3d
3. image      2h         400         33d
4. image      1d         400        400d

In my point of view, this will be only worthfull if the *impression*
those images gives about the amount of throughput is about the real
throuput (I wonder what the notorius "Bytes per Second" should mean with
an graph showing an green area of approx 3 qcm around 18 o'clock ;-).

Maybe you want to have an image No 0 showing only last hour/last few hours
to get better control over recent happenings. This will look stairy, but
now problem. It's can even run from the same database.


I will give the above numbers a chance and will send you a note, when i've
broke it down to an easy to use script.

cheers
Michael