[Users] slurm

[Francisco Guzman]

Thanks a lot Erik and Ian.

Francisco

----- Original Message -----
From: "Ian Hinder" <ian.hinder at aei.mpg.de>
To: "Francisco Guzman" <guzman at ifm.umich.mx>
Cc: "Einstein Toolkit Users" <users at einsteintoolkit.org>
Sent: Friday, March 3, 2017 2:52:28 AM
Subject: Re: [Users] slurm

On 2 Mar 2017, at 20:28, Francisco Guzman <guzman at ifm.umich.mx> wrote:

> 
> Hi all,
> 
> Does anyone have experience using the queue administrator SLURM when submitting ETK jobs?
> 
> Particularly important would be to know how specific one has to be at declaring MODULES within the submission-job scripts, which require specific library and tools PATHS.
> 
> I'd appreciate if someone has a script that works (and can pass it on to me).

Hi Francisco,

SLURM is used on many systems with the ET, and seems to be the standard for new systems.

The choice of modules that are needed on a system is more or less orthogonal to the choice of queuing system (SLURM).  I don't think there is anything special you have to worry about in the interaction between the ET and SLURM; just specify the number of nodes etc as you would for any other code.  

For loading modules and setting paths in the optionlist, it is usually quite a bit of work to get the ET running on a new system, and there's not a lot that can be done about that, since every system is different.

There are example SLURM submission scripts in SimFactory, as Erik said.  You can look in 

	https://bitbucket.org/simfactory/simfactory2/src/master/mdb/submitscripts/?at=master

for example at Comet or Minerva, which both use SLURM.

The optionlists are in https://bitbucket.org/simfactory/simfactory2/src/master/mdb/optionlists/?at=master, but they are specific to the clusters, and shouldn't care whether they are running under SLURM or not.

-- 
Ian Hinder
http://members.aei.mpg.de/ianhin