HPC/Submitting and Managing Jobs/Example Job Script - Revision history

Stern: /* Job scripts in other scripting languages */

2019-10-08T22:09:44Z

Job scripts in other scripting languages

← Older revision		Revision as of 22:09, October 8, 2019
Line 121:		Line 121:

	Then, a job script in Python would typically begin like this:		Then, a job script in Python would typically begin like this:
	#!~~'''~~/usr/bin/env python3~~'''~~		<source lang="python3">
	#PBS -l nodes=''N'':ppn=''PPN''		#!/usr/bin/env python3
	#PBS -l walltime=''hh:mm:ss''		#PBS -l nodes=N:ppn=PPN
	#PBS -N ''job_name''		#PBS -l walltime=hh:mm:ss
	#PBS -A ''account''		#PBS -N job_name
	#PBS -o ''filename''		#PBS -A account
	#PBS -j oe		#PBS -o filename
	#PBS -m ea		#PBS -j oe
			#PBS -m ea
	~~import os~~
	~~if "PBS_O_WORKDIR" in os.environ:~~
	~~'''os.chdir(os.environ["PBS_O_WORKDIR"])'''~~

	~~# body of script here …~~
	'''Important:''' The job script only ever runs in serial, on one core. Any parallelization must be initiated by leveraging MPI or OpenMP explicitly (e.g. http://mpi4py.scipy.org) or implicitly by calling a child application that runs in parallel (e.g. VASP). Such a child script or application either needs to be aware of or be told to use <code>$PBS_NODEFILE</code>.

	This is not to be confused with using ~~a "regular" bash script and~~ mpirun to start several ~~custom python~~ interpreters in parallel, which is how [[HPC/Modules/gpaw \| GPAW]] and [[HPC/Modules/atk \| ATK]] operate:		import os
			if "PBS_O_WORKDIR" in os.environ:
			os.chdir(os.environ["PBS_O_WORKDIR"])

			# body of script here …
			</source>

			'''Important:''' The job script ''interpreter'' only ever runs in serial, on one core.
			Any parallelization must be initiated by one of the following mechanisms:

			* Calling a child application that runs in parallel (e.g. VASP). Such a child script or application either needs to be aware of or be told to use <code>$PBS_NODEFILE</code>. When the child application finishes, the calling script will collect results and typically iterates until some criterion is met. This calling model is perfectly reasonable as long as the "controlling" application doesn't dominate the time required to shepherd its child calculations.
			* OpenMP
			* [https://docs.python.org/3/library/multiprocessing.html Python <code>multiprocessing</code>]
			* MPI (e.g. [https://docs.python.org/3/library/multiprocessing.html Mpi4Py])

			The latter is not to be confused with using mpirun to start several Python interpreters in parallel, which is how [[HPC/Modules/gpaw \| GPAW]] and [[HPC/Modules/atk \| ATK]] operate:
	#!'''/bin/bash'''		#!'''/bin/bash'''
	…		…

Stern: /* Job scripts in other scripting languages */

2019-10-08T21:55:49Z

Job scripts in other scripting languages

← Older revision		Revision as of 21:55, October 8, 2019
Line 121:		Line 121:

	Then, a job script in Python would typically begin like this:		Then, a job script in Python would typically begin like this:
	#!'''/usr/bin/~~python~~'''		#!'''/usr/bin/env python3'''
	#PBS -l nodes=''N'':ppn=''PPN''		#PBS -l nodes=''N'':ppn=''PPN''
	#PBS -l walltime=''hh:mm:ss''		#PBS -l walltime=''hh:mm:ss''

Stern: /* Job scripts in other scripting languages */

2019-06-19T16:26:32Z

Job scripts in other scripting languages

← Older revision		Revision as of 16:26, June 19, 2019
Line 131:		Line 131:

	import os		import os
	if 'PBS_O_WORKDIR' in os.environ:		if "PBS_O_WORKDIR" in os.environ:
	'''os.chdir(os.environ['PBS_O_WORKDIR'])'''		'''os.chdir(os.environ["PBS_O_WORKDIR"])'''

	# body of script here …		# body of script here …

Stern: /* Job scripts in other scripting languages */

2019-06-19T16:26:02Z

Job scripts in other scripting languages

← Older revision		Revision as of 16:26, June 19, 2019
Line 131:		Line 131:

	import os		import os
	if os.environ~~.has_key('PBS_O_WORKDIR')~~:		if 'PBS_O_WORKDIR' in os.environ:
	'''os.chdir(os.environ['PBS_O_WORKDIR'])'''		'''os.chdir(os.environ['PBS_O_WORKDIR'])'''

Stern: /* Timed Allocations */

2019-04-12T04:01:07Z

Timed Allocations

← Older revision		Revision as of 04:01, April 12, 2019
Line 166:		Line 166:

	=== Timed Allocations ===		=== Timed Allocations ===
	See [[HPC/FAQ#My account balance ~~suddenly dropped~~]].		See [[HPC/FAQ#What's my account balance?]].

	== Advanced node selection ==		== Advanced node selection ==

Stern: /* Timed Allocations */

2015-08-24T17:33:19Z

Timed Allocations

@@ Line 166: / Line 166: @@
 === Timed Allocations ===
-The compute time physically available by Carbon's processors is a perishable resource.
+See [[HPC/FAQ#My account balance suddenly dropped]].
 == Advanced node selection ==

Stern: /* Job scripts in other scripting languages */

2013-06-19T21:02:28Z

Job scripts in other scripting languages

← Older revision		Revision as of 21:02, June 19, 2013
Line 133:		Line 133:
	if os.environ.has_key('PBS_O_WORKDIR'):		if os.environ.has_key('PBS_O_WORKDIR'):
	'''os.chdir(os.environ['PBS_O_WORKDIR'])'''		'''os.chdir(os.environ['PBS_O_WORKDIR'])'''
	…
			# body of script here …
	'''Important:''' The job script only ever runs in serial, on one core. Any parallelization must be initiated by leveraging MPI or OpenMP explicitly (e.g. http://mpi4py.scipy.org) or implicitly by calling a child application that runs in parallel (e.g. VASP). Such a child script or application either needs to be aware of or be told to use <code>$PBS_NODEFILE</code>.		'''Important:''' The job script only ever runs in serial, on one core. Any parallelization must be initiated by leveraging MPI or OpenMP explicitly (e.g. http://mpi4py.scipy.org) or implicitly by calling a child application that runs in parallel (e.g. VASP). Such a child script or application either needs to be aware of or be told to use <code>$PBS_NODEFILE</code>.

Stern: /* Advanced Job scripts */

2013-01-28T21:30:43Z

Advanced Job scripts

← Older revision		Revision as of 21:30, January 28, 2013
Line 100:		Line 100:

	== Advanced Job scripts ==		== Advanced Job scripts ==
	=== ~~Multi-threading under~~ OpenMP ===		=== Multithreading using OpenMP ===
	See [[HPC/Submitting and Managing Jobs/Advanced node selection#Multithreading]].		See [[HPC/Submitting and Managing Jobs/Advanced node selection#Multithreading using OpenMP]].

	=== GPU nodes ===		=== GPU nodes ===

Stern: /* GPU nodes */

2013-01-28T21:29:46Z

GPU nodes

← Older revision		Revision as of 21:29, January 28, 2013
Line 98:		Line 98:
	programname		programname
	</syntaxhighlight>		</syntaxhighlight>

			== Advanced Job scripts ==
			=== Multi-threading under OpenMP ===
			See [[HPC/Submitting and Managing Jobs/Advanced node selection#Multithreading]].

	=== GPU nodes ===		=== GPU nodes ===

Stern: /* GPU nodes */

2012-11-14T23:33:44Z

GPU nodes

← Older revision		Revision as of 23:33, November 14, 2012
Line 100:		Line 100:

	=== GPU nodes ===		=== GPU nodes ===
	As of late 2012, Carbon has 28 nodes with 1 NVIDIA C2075 GPU each. To specifically request GPU nodes, add <code>:gpus=1</code> to the <code>nodes</code> request:		As of late 2012, Carbon has 28 nodes with 1 NVIDIA C2075 GPU each. To specifically request GPU nodes, add <code>:gpus=1</code> to the <code>nodes</code> request. At the moment this is synonymous with but preferable to specifying <code>…:gen3</code>.
	#PBS -l nodes=''N'':ppn=''PPN''''':gpus=1'''		#PBS -l nodes=''N'':ppn=''PPN''''':gpus=1'''
	~~At the moment this is synonymous with but preferable to:~~
	~~#PBS -l nodes=…:gen3~~
	* The <code>gpus=…</code> modifier refers to GPUs ''per node'' and currently must always be 1.		* The <code>gpus=…</code> modifier refers to GPUs ''per node'' and currently must always be 1.
	* GPU support depends on the application, in particular if several MPI processes or OpenMP threads can share a GPU. Test and optimize the <code>''N''</code> and <code>''PPN''</code> parameters for your situation. Start with <code>nodes=1:ppn=4</code>.		* GPU support depends on the application, in particular if several MPI processes or OpenMP threads can share a GPU. Test and optimize the <code>''N''</code> and <code>''PPN''</code> parameters for your situation. Start with <code>nodes=1:ppn=4</code>.