Discussion group for the ARIA software

[Benjamin Bardiaux <bardiaux AT pasteur.fr>]

Dear Benedikt,

My bad, I forgot to mention the required "—no-test" option in the aria2 command line when using sbatch submission.

(It will disable the host checking).

Best,

Benjamin 

-----------------------------------------------------
Dr Benjamin Bardiaux    | Bioinformatique Structurale
bardiaux AT pasteur.fr     | Institut Pasteur
25,28 rue du Docteur Roux 75015 Paris, France
-----------------------------------------------------

Le 3 mai 2021 à 10:43, Benedikt Soeldner <benedikt.soeldner AT tu-dortmund.de> a écrit :

Dear Benjamin,

thank you for the tip. I tried it out on friday, but unfortunately,
already at the beginning, the following error occured:

MESSAGE [Project]: Checking host list ...
WARNING [Job manager]: Command "sbatch --job-name=CNS_Aria_run_2021-04-30
                      --error=/.../cns_error_msg.txt --output=/.../cns_
                      output_msg.txt --partition=short --ntasks=1
                      --cpus-per-task=1" ... failed (connection failed or
                      unable to open temporary file.)
WARNING [Project]: At least one command could not be executed. Please check
                  your host list setup.

Is the reason for this problem maybe, that the check_host.csh file in the
temporary direction doesn't look like a Slurm Slurm job script (in
contrast to the files refine.csh and refine_water.csh in the temporary
directory)? If yes, does this file needs to be modified or is there
another problem?

Best regards,

Benedikt

Dear Benedikt,

If you're slurm setup allows submission from the node where you're running

aria, you could use

      <host enabled="yes" command="sbatch --your.sbatch.options"

executable="/.../programs/cns_solve_1.21/intel-x86_64bit-linux/bin/cns_solve"

n_cpu="20" use_absolute_path="yes"/>

    </job_manager>

this means that each CNS calculation will be submitted independently via

sbatch, hence running in parallel.

Best regards,

Benjamin

On 30/04/2021 11:19, Benedikt Soeldner wrote:

Dear Aria discussion group,

since a few months, I'm running my Aria calculations on a cluster

managed

with the Slurm scheduling system. Unfortunately, I didn't find out yet,

how to run Aria, i.e., the CNS calculation processes, on multiple nodes

in

parallel. Does anyone of you know if and how this works?

At the moment, I start my Aria calculations by submitting a job script,

which looks like the following:

#!/bin/bash -l

#SBATCH --job-name=Aria-Project_run01

#SBATCH --output=Aria-Project_run01.out

#SBATCH --error=Aria-Project_run01.err

#SBATCH --partition=short

#SBATCH --ntasks=1

#SBATCH --cpus-per-task=20

#SBATCH --mem=8G

module purge

module load python/2.7.18

cd /.../Aria-Project

srun python -O /.../programs/aria2.3.2/aria2.py --output=run01msg.txt

Aria-Project_run01.xml

And in the Aria project file, the following lines are written:

    <job_manager default_command="csh -f">

      <host enabled="yes" command="csh -f"

executable="/.../programs/cns_solve_1.21/intel-x86_64bit-linux/bin/cns_solve"

n_cpu="20" use_absolute_path="yes"/>

    </job_manager>

When I increase the number of nodes (#SBATCH --nodes=... in the job

script), the job still runs just on one node, as there is only one

initial

task. I guess, the lines above from the project file also need to be

modified somehow and not only in the Slurm job script. Do you know, how

these two files should look like? Or do I need to modify other files in

addition?

Thank you for your help!

Best regards,

Benedikt Söldner

--------------------------------------

Benedikt Söldner (PhD student)

benedikt.soeldner AT tu-dortmund.de

Technical University Dortmund, Germany

Research group of Prof. Dr. Rasmus Linser

--

---------------------------------------------

Dr Benjamin Bardiaux      bardiaux AT pasteur.fr

Unité de Bioinformatique Structurale

CNRS UMR3528 - Institut Pasteur

25,28 rue du Docteur Roux 75015 Paris, France

---------------------------------------------