[IPython-dev] ipcluster and PBS engines

[Antonio González Peña]

Thank you for your help.

On Mon, Nov 3, 2014 at 1:22 PM, MinRK <benjaminrk at gmail.com> wrote:
> It’s up to your PBS system how it assigns the individual jobs in a job
> array. Under normal circumstance, this will be spread across nodes. If it
> isn’t you may need to use custom PBS templates to assign jobs how you want,
> or submit each bunch of engines separately:
>
> for i in {1..$NODES}; do
>   ipcluster engines -n $ENGINES_PER_NODE &
> done
>
> The default job file looks like:
>
> #!/bin/sh
> #PBS -t 1-10
> #PBS -V
> #PBS -N ipengine
> /usr/bin/python -m IPython.parallel.engine
>
> -MinRK
>
>
> On Mon, Nov 3, 2014 at 12:11 PM, Antonio González Peña <antgonza at gmail.com>
> wrote:
>>
>> Just looking at qstat as I was expecting to have 10 actual jobs
>> running. Expanding on that, what I was expecting is that if I have,
>> let's say, 32 cores (4 nodes/workers) in all my cluster and I specify
>> --n 10 that I have 10 engines and they are distributed between the 4
>> nodes. Is this possible?
>>
>> On Mon, Nov 3, 2014 at 12:35 PM, MinRK <benjaminrk at gmail.com> wrote:
>> > It should create one PBS job array that starts ten engines. Are you only
>> > looking at qstat, or are you looking at the actual number of engines
>> > that
>> > end up connected via a parallel.Client?
>> >
>> > -MinRK
>> >
>> >
>> > On Mon, Nov 3, 2014 at 10:17 AM, Antonio González Peña
>> > <antgonza at gmail.com>
>> > wrote:
>> >>
>> >> Hi,
>> >>
>> >> Let's say that I have a system with n nodes and m cores and I will
>> >> like to start 10 engines using PBS. Note that n*m>10 and all my
>> >> filesystems are shared so that's not a problem. My current issue is
>> >> that if do:
>> >> ipcluster start --n 10
>> >> and/or have in my ipcluster_config.py
>> >> c.IPClusterEngines.n = 10
>> >> I always get 1 controller and 1 engine vs. having 1 controller and 10
>> >> engines submitted to PBS.
>> >>
>> >> Other option is that I'm misunderstanding how this works.
>> >>
>> >> Any help will be greatly appreciated.
>> >>
>> >> --
>> >> Antonio
>> >> _______________________________________________
>> >> IPython-dev mailing list
>> >> IPython-dev at scipy.org
>> >> http://mail.scipy.org/mailman/listinfo/ipython-dev
>> >
>> >
>> >
>> > _______________________________________________
>> > IPython-dev mailing list
>> > IPython-dev at scipy.org
>> > http://mail.scipy.org/mailman/listinfo/ipython-dev
>> >
>>
>>
>>
>> --
>> Antonio
>> _______________________________________________
>> IPython-dev mailing list
>> IPython-dev at scipy.org
>> http://mail.scipy.org/mailman/listinfo/ipython-dev
>
>
>
> _______________________________________________
> IPython-dev mailing list
> IPython-dev at scipy.org
> http://mail.scipy.org/mailman/listinfo/ipython-dev
>



-- 
Antonio