Re: [Condor-users] Requirements not respected

[Steffen Grunewald <steffen.grunewald@xxxxxxxxxx>]

On Wed, May 23, 2007 at 03:50:07PM -0400, Fr?d?ric Bastien wrote:
> Hi Thanks for the information. I don't know if it is a mistake but 
> TotalCpus is not in the manual in section 2.5.2.1. Meaby someone could 
> add it?

Oh, the manual... indeed, there are config items which are mentioned only
in the "prose" part of the manual but not the config reference, and
other you may only find out about by closely reading the output of condor_status.

> This solve partially my trouble as this allow the jobs to be executed on 
> the correcte computers. But if my programme need 4 cpu on the same 
> machine and I do ( target.TotalCpus == 4 ), it don't reserve 4 cpus on 
> this machine for my job. I have checked the parallele universe but it 
> don't seem to by the good approch as this is more parallele job between 
> multiple machine.

If you put TotalCpus == 4 into your Requirements, this will only match 
VMs that are located on a multi-CPU machine. Still matching is done per
VM, currently there's no way to match to multiple VMs. 
Peter's approach - though looking ugly IMHO - works around this by defining 
"special VMs", and does it right (by counting the number of free CPUs and !STARTing
the VMs which are no longer available). Very nice, but still a kludge. (Here
we found the point where Condor isn't PBS)

> Do someone know how to reserve multiple CPU or VM on one machine for one 
> job(One cpu by VM)? Is this the same probleme with dynamic memory 
> allocation that passed recently on the mailing list?

Memory allocation might be done in a very similar way (but memory footprints
change more dynamically, and you might run into swap). Volunteers welcome! :-)

Steffen

-- 
Steffen Grunewald * MPI Grav.Phys.(AEI) * Am Mühlenberg 1, D-14476 Potsdam
Cluster Admin * http://pandora.aei.mpg.de/merlin/ * http://www.aei.mpg.de/
* e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
No Word/PPT mails - http://www.gnu.org/philosophy/no-word-attachments.html