RES: [Orca-users] Creating CPU graphs based on processor sets.

[Kevin Smolkowski]

This is running on a F12 so we could run multiple domains
if we wanted.   (And we do for some applications).  For this
project, we separate our databases and applications to keep 
them somewhat isolated, but we like the ease of keeping them 
on the same host.   Also processor sets are easy to change 
around on the fly.   Since nobody else has done it, I'll take 
a stab at using the SE toolkit to extract the data.

Thanks for the responses!

Quoting Guilherme Carvalho Chehab <Guilherme.Chehab at tco.net.br>:

> Kevin,
> 
> I think that even you can manage to have many orcallator.se´s running at
> same time I would guess that they are still going to gather data of all
> processors instead of the amount of total cpus for their respective
> processor sets. Because as there is still only one kernel running on the
> machine, so the performance data of the total cpus usage would have the same
> reading from the SE package.
> 
> More than that, all other data as disk, network and memory usage would be
> collected with redundance.
> 
> This way, if your machine support domains like E10k ou the new sunfire
> servers you could achieve the same result by dividing the load in different
> domains, each on with its own running kernel. That is the way we solved the
> problem of having the load distributed between various database instances
> 
> But if your need is really to work on processor sets - or your servers don´t
> support domains distributions (or you don´t want to bother messing with it).
> You can try to tweak orcallator.se to collect data from each CPU - I think
> that it is possible to do with SE package - and combine these information
> with the defined processor sets - I would bet that that also is possible
> with SE toolkit. I believe that you can even a count on how many processes
> are bound to each processor set... :-)
> 
> Each information you collect has to be outputed as a new column on the
> output files from orcallator.se. And inserted as a plot on the
> orcallator.cfg.
> 
> I did a grep -i "procesor set" on /opt/RICHPse/include on my SE 3.2 and have
> found hits on proc.se and proccess_class.se. It where I would look for
> information on how to do it.
> 
> Even so, If SE doesn´t have this data easily available, you can include on
> orcallator.se a external program call to "mpstat" and to "psrset -l"
> redirecting their outputs for temporary files and treat it in orcallator.se
> (is almost like programming in plain C) inserting them on the output files.
> 
> Although it is a dirty way to do it, I did something like these to call
> veritas volume manager information, and works pretty well.
> 
> 
> Regards,
> Guilherme
> 
> > -----Mensagem original-----
> > De: Blair Zajac [mailto:blair at orcaware.com]
> > Enviada em: quarta-feira, 8 de janeiro de 2003 18:31
> > Para: Kevin Smolkowski
> > Cc: orca-users at orcaware.com
> > Assunto: Re: [Orca-users] Creating CPU graphs based on processor sets.
> >
> >
> > Kevin Smolkowski wrote:
> > >
> > > We make use of processor sets to divide up the workload between
> > > various instances of our database engine.
> > >
> > > Has anyone modified Orca to work with processor sets?
> >
> > Kevin,
> >
> > No, there are no modifications for this.  I haven't worked on
> > a system with processor sets used, but maybe you could run
> > separate orcallator.se's on each one and have Orca treat them
> > as separate hosts.
> >
> > Best,
> > Blair
> >
> > --
> > Blair Zajac <blair at orcaware.com>
> > Plots of your system's performance -
> > http://www.orcaware.com/orca/
> > _______________________________________________
> > Orca-users mailing list
> > Orca-users at orcaware.com
> > http://www.orcaware.com/mailman/listinfo/orca-> users
> >
>