[StarCluster] SGE have a job consume all slots?

Justin Riley jtriley at MIT.EDU
Fri Feb 24 10:36:20 EST 2012


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Correct me if I'm wrong but if you modify the 'orte' parallel
environment on StarCluster to use the '$pe_slots' allocation_rule:

$ qconf -mp orte
allocation_rule $pe_slots

then according to the SGE docs:

"If the special denominator $pe_slots is used, the full range of
processes as specified with the qsub(1) -pe switch has to be allocated
on a single host (no matter which value belonging to the range is
finally chosen for the job to be allocated)."

So from my understanding after making the above change the following
would use all 8 slots on a single machine for a single job:

$ qsub -cwd -pe orte 8 ./jobscript.sh

HTH,

~Justin

On 02/23/2012 09:43 PM, Rayson Ho wrote:
> If you want every job (OpenMP job?) to consume all the CPUs, then 
> forcing all nodes to have a slot of 1 would work.
> 
> But if you have a mix of serial and threaded workloads, then the
> correct way of ensuring exclusive execution is still via SGE's
> built-in mechanism.
> 
> Rayson
> 
> ================================= Open Grid Scheduler / Grid
> Engine http://gridscheduler.sourceforge.net/
> 
> Scalable Grid Engine Support Program http://www.scalablelogic.com/
> 
> 
> ------------------------------------------------------------------------
>
> 
*From:* Don MacMillen <macd at nimbic.com>
> *To:* David Erickson <derickso at stanford.edu> *Cc:*
> starcluster at mit.edu *Sent:* Thursday, February 23, 2012 8:43 PM 
> *Subject:* Re: [StarCluster] SGE have a job consume all slots?
> 
> Hmm, the docs don't seem to be all that enlightening.  Here is
> another way. init_node is called for every node in the run method
> of a starcluster plugin as well as in the on_add_node method of the
> plugin. HTH.
> 
> Regards,
> 
> Don
> 
> 
> def init_node(self, node, master): ... # Set the number of slots to
> 1.  We do this so that only one job is # is submitted per machine,
> since we will use all of its available # threads in multi-threaded
> mode. cmd_strg = 'source /opt/sge6/default/common/settings.sh;' \ 
> 'qconf -mattr exechost complex_values slots=1 %s' % node.alias
> 
> self.logger.debug("Executing: |%s|" % cmd_strg) output =
> master.ssh.execute(cmd_strg) ...
> 
> 
> 
> 
> On Thu, Feb 23, 2012 at 3:56 PM, David Erickson
> <derickso at stanford.edu <mailto:derickso at stanford.edu>> wrote:
> 
> Hi- I've been digging through SGE/OGS docs for the last hour or so
> trying to sort out the easiest way to enforce a one job per host
> restriction, does anyone have a suggestion on how to do this?  My
> hosts are coming up with 8 slots, so I tried launching with -l
> slots=8 but it complained about wanting me to use parallel
> environments which looks even more complicated..
> 
> Thanks, David _______________________________________________ 
> StarCluster mailing list StarCluster at mit.edu
> <mailto:StarCluster at mit.edu> 
> http://mailman.mit.edu/mailman/listinfo/starcluster
> 
> 
> 
> _______________________________________________ StarCluster mailing
> list StarCluster at mit.edu <mailto:StarCluster at mit.edu> 
> http://mailman.mit.edu/mailman/listinfo/starcluster
> 
> 
> 
> 
> This body part will be downloaded on demand.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk9HrnMACgkQ4llAkMfDcrl5PQCfXDvnnLz1l+M0bQur55YbgNpe
/gMAniaF2s01JtHL0qJdNrIjr93udhJg
=ZQKm
-----END PGP SIGNATURE-----


More information about the StarCluster mailing list