[Orca-users] Orcallator.cfg: Add threshold for # of cpus to Average # Processesim Run Queue-Graph

[Sean O'Neill]

At 02:25 PM 2/24/2003 +0100, B.Schopp at gmx.de wrote:
> > B.Schopp at gmx.de wrote:
> > >
> > > Hi all,
> > >
> > > i just have seen within the datafiles generaterd by orcallator, that
> > the
> > > number of cpus is also being monitored. As i missed the value within
> > > orca i added it to orcallator.cfg into the 'Average # Processes in Run
> > > Queue'-Graph (like in procallator i set the line for the # of cpus to
> > > line2).
> >
> > I generated this plot for some of the data I have for a 2 CPU box, and for
> > this box, it gets very little usage so that the actual usage plots don't
> > show a lot of detail.
> >
> > I've attached a sample plot.
>
>So did i... ;-)
>
> >
> > So in my mind, this doesn't show too much detail in the usage.  I would
> > like to see more detail in the actual usage then just see a plot scaled
> > to the number of CPUs.
>
> >From my point of view, the values below 0.1 aren't really of any interest
>on behalf of the average load. On the other side, it may be interesting if
>you are developing and want to countercheck the performance of a script
>or a new binary. On active (and busy) servers the # of cpus will be a point
>of interest because of the possibilty to check out if the system has enough
>cpus...
>
> >
> > However, I am willing to consider adding this to the official distribution
> > if a majority of people vote to see it added.
> >
> > > What do you think about the colours: The line for the # of  cpus gets as
> > > colour magenta by default. Should it be changed ilike in procallator?
> >
> > Regarding the colors, I'd like them all to be consistent between
> > orcallator
> > and procallator.  Can you send a sample procallator plot so I can compare
> > color schemes?  As a note, I've been treating orcallator as the definitive
> > source.
>
>The difference in the colours is that in procallator the line colour for the
># of cpus is red and in orcallator it is magenta. From a logical point of
>view,
>the red line could be seen as an alarm signal if it crossed by the load.
>I agree that the colours should be consistent within the same graphs.

Now that I'm thinking about this - The words "load threshold" really aren't 
good words to use here.  Just call it exactly what it is - a CPU count.

Just because a 6 CPU system has a 1 minute or even 15 minute load average 
of 6 really means nothing from a performance standpoint in 99.9% of the 
cases.  It's when that load average crosses the 2x CPU count (or even 
3x-4x) that you need to start asking "Is there a problem here ?"

Every environment is going to be different so this 2x-CPU-average-load rule 
is a gross-average for every environment and needs to be carefully 
re-qualified.

I'm probably splitting hairs but this is just a "IMHO" ...


--
........................................................
......... ..- -. .. -..- .-. ..- .-.. . ... ............
.-- .. -. -... .-.. --- .-- ... -.. .-. --- --- .-.. ...

Sean O'Neill