Python Module paralleltools

Paralleltools.py Module to data-parallelize simulations and write pbs scripts. See the manual for examples of parallel usage.

last revision 070313-MLW

paralleltools.Writepbsscript(fileStub, nodes, ExecutableName='Execute_MPSParallelMain', InputFileNames=[], MyPyFile=None, time='12:00:00', computer='mio', myusername=None)

Write a pbs script for the given parallel job. Returns name of the pbs script written.

Arguments

fileStubstr

the job id to identify the simulation

nodesstr (maybe int, less likely list of nodes???)

specifying the number of nodes on the cluster.

ExecutableNamestr

name of the OpenMPS parallel executable name default to ExecuteMPSParallelMain

InputFileNameslist

list what to execute default to empty list

MyPyFilestr

corresponding python script copied to keep track of the input for the job. default to None

timestr

walltime for the simulation. default to 12:00:00

computerstr

Choosing between HPC mio and ra. default to mio

myusernamestr

used to email information about the job status to your @mines.edu email adress. default to None

paralleltools.WriteMPSParallelFiles(Parameters, Operators, HamiltonianMPO, comp_info, PostProcess=False)

Write MPI files for fortran. Returns either name of the pbs/slurm script for PostProcess=False or the dictionaries with simulations (being consistent with WriteMPSParallelTemplate).

Arguments

Parameterslist of dictionaries

contains the simulation for each simulation in dictionaries.

Operatorsdictionary

containing all operators necessary for the simulation.

HamiltonianMPOinstance of MPO.MPO

defines the Hamiltonian of the system.

comp_infodictionary

information for setting up a simulation on the cluster. Key queueing decides between using pbs (pbs) or slurm/sbatch (slurm) format. If key not given, the default key pbs for pbs is set automatically. Keys are for the pbs-script computer (default mio), myusername (None, used for email), time (12:00:00), nodes (obligatory), and ThisFileName (None, uses then random string). For the slurm/sbatch please view the function description of the dictionary cinfo in Paralleltools.write_sbatch_script().

PostProcessBool, optional

flag if simulation should be run (False) or if only data is analyzed (True). default to False

paralleltools.WriteMPSParallelTemplate(ptemplate, Operators, H, comp_info, staticParamsforIteration=[], staticIterators=[], itermethod='CartesianProduct', PostProcess=False)

Write parallel files that the master can send to worker units. Returns list or dictionary with simulation settings for PostProcess=True, and otherwise the name of the pbs/slurm script.

Arguments

ptemplatedictionary

dictionary containing the simulation settings

Operatorsdictionary

containing all operators necessary for the simulation.

Hinstance of MPO.MPO

defines the Hamiltonian of the system.

comp_infodictionary

information for setting up a simulation on the cluster. Key queueing decides between using pbs (pbs) or slurm/sbatch (slurm) format. If key not given, the default key pbs for pbs is set automatically. Keys are for the pbs-script computer (default mio), myusername (None, used for email), time (12:00:00), nodes (obligatory), and ThisFileName (None, uses then random string). For the slurm/sbatch please view the function description of the dictionary cinfo in Paralleltools.write_sbatch_script().

staticParamsforIterationlist, optional

parameters to be iterated over given through their identifier string. default to empty list

staticIteratorslist, optional

corresponding values for the parameters to be iterated over. Order should correspond to the order in staticParamsforIteration. default to empty list

itermethodstring

Defining how to derive the simulation parameters from the list of iterators. Options are 1) CartesianProduct, e.g. building from parameters values [a1, a2] and [b1, b2] the four simulations (a1, b1), (a1, b2), (a2, b1) and (a2, b2). 2) Linear, e.g. building from the parameter values [a1, a2] and [b1, b2] the two simulations (a1, b1) and (a2, b2). default to CartesianProduct

PostProcessBool, optional

flag if simulation should be run (False) or if only data is analyzed (True). default to False

paralleltools.WriteParallelfiles(commands, job_ID, writedir='', outdir='')

Write parallel files from Params

Arguments

commandslist

containing the main setting files passed to fortran for each simulation, e.g. as returned by tools.WriteFiles().

job_IDstr

job identifier of the parameter dictionary.

writedirstr, optional

could the temporary directory of the MPS simulation specified in the parameter dictionary default to empyt string

outdirstr, optional

could be the output directory specified in the parameter dictionary. default to empty string

paralleltools.write_sbatch_script(fileStub, cinfo, ExecutableName='Execute_MPSParallelMain', InputFileNames=[])

Write a slurm/sbatch script for submitting a job on a cluster. Returns name of slurm script written.

Arguments

fileStubstr

the job id to identify the simulation. If you run multiple simulation this one must already be unique at the current stage

cinfodictionary

containing all the setting for the cluster. The following keys can defined:

• myusername : used to email information about the job status to your @mines.edu email adress. Default to None.

• computer : used to deduct the number of cores. More about the the number of tasks specified you can find in Paralleltools.get_ntasks(). Default to mio.

• time : walltime for the simulation, default to 12:00:00.

• nodes : can be an integer with the number of nodes or a list with three-character string specifying the nodes. Default to 1.

• ThisFileName : corresponding python script copied to keep track of the input for the job. Default to None.

• partition : set partition flag of SBATCH, default to None.

• mem : set memory for job, default to None.

• exclusive : if True, nodes are blocked for any other jobs in case any remaining cores would be available. Default to True.

• mpi : choose mpi executer, default to srun (alternative mpiexec)

ExecutableNamestr

name of the OpenMPS parallel executable name default to ExecuteMPSParallelMain

InputFileNameslist

list what to execute default to empty list

paralleltools.get_ntasks(cinfo)

Return the number of nodes as string based on a nodelist or on the number of nodes. If no information is available, eight cores per node are assumed.

Arguments

cinfodictionary

containing the setting of the job for a cluster, used within here are the dictionary keys computer and nodes

paralleltools.get_nodelist(cinfo)

Return the nodelist as string

Arguments

cinfodictionary

containing the setting of the job for a cluster, used within here is the dictionary key nodes which should be a list when calling this function.

paralleltools.WriteFiles_mpi(Parameters, Operators, HamiltonianMPO, PostProcess=False)

This files has the same purpose as tools.WriteFiles() for MPI runs with mpi4py and the argument description should be read there. The files are only written once and broadcasted to every core.

paralleltools.WriteFiles_mpi(Parameters, Operators, HamiltonianMPO, PostProcess=False)

This files has the same purpose as tools.WriteFiles() for MPI runs with mpi4py and the argument description should be read there. The files are only written once and broadcasted to every core.

paralleltools.encode_list(mylist)

Put a list into a dictionary where the key is the position (as string) of the entry.

Arguments

mylistlist

is converted to the described dictionary format.

paralleltools.encode_list(mylist)

Put a list into a dictionary where the key is the position (as string) of the entry.

Arguments

mylistlist

is converted to the described dictionary format.

paralleltools.decode_list(mydic)

Decode a dictionary from Paralleltools.encode_list() back into a list.

Arguments

mydicdictionary

dictionary in the corresponding form, keys are 0 to lenght - 1 (as integer)

paralleltools.decode_list(mydic)

Decode a dictionary from Paralleltools.encode_list() back into a list.

Arguments

mydicdictionary

dictionary in the corresponding form, keys are 0 to lenght - 1 (as integer)

paralleltools.runMPS_mpi(infileList, RunDir=None, customAppl=None)

This file has the same purpose as tools.runMPS() for MPI runs with mpi4py and the arguments are restricted. You have to call the python script with the command line argument --mpi4py to load mpi4py modules.

Arguments

infileListlist

list of of the main files passed to fortran executable as command line argument.

RunDirstr, optional

Specifies path of the fortran-executable (local or global installation). This is available for default-simulation and custom application. (@Developers: Debugging with valgrind always works with a local installation.) None : if executable available within local folder, use local installation, otherwise assume global installation, default ‘’ (empty string) : for global installation ‘./’ : local installation PATH+’/’ : path to executable ending on slash /. Additional steps may be necessary when using this.

customApplstr, optional

define custom executable. Global and local installation work as before. Custom applications cannot run with valgrind. default to None (running default executable)

paralleltools.runMPS_mpi(infileList, RunDir=None, customAppl=None)

This file has the same purpose as tools.runMPS() for MPI runs with mpi4py and the arguments are restricted. You have to call the python script with the command line argument --mpi4py to load mpi4py modules.

Arguments

infileListlist

list of of the main files passed to fortran executable as command line argument.

RunDirstr, optional

Specifies path of the fortran-executable (local or global installation). This is available for default-simulation and custom application. (@Developers: Debugging with valgrind always works with a local installation.) None : if executable available within local folder, use local installation, otherwise assume global installation, default ‘’ (empty string) : for global installation ‘./’ : local installation PATH+’/’ : path to executable ending on slash /. Additional steps may be necessary when using this.

customApplstr, optional

define custom executable. Global and local installation work as before. Custom applications cannot run with valgrind. default to None (running default executable)

paralleltools.master_mpi(comm, ncores, rank, infileList, RunDir, customAppl)

This is the master for mpi runs taking care of the work distribution and calculating jobs via threading.

Arguments

comminstance of mpi4py.MPI.COMM_WORLD

communicator for the MPI program

ncoresint

number of processes in the MPI-run

rankint

rank of this process (should be 0)

infileListlist

list of of the main files passed to fortran executable as command line argument.

RunDirstr, optional

Specifies path of the fortran-executable (local or global installation). This is available for default-simulation and custom application. (@Developers: Debugging with valgrind always works with a local installation.) None : if executable available within local folder, use local installation, otherwise assume global installation, default ‘’ (empty string) : for global installation ‘./’ : local installation PATH+’/’ : path to executable ending on slash /. Additional steps may be necessary when using this.

customApplstr, optional

define custom executable. Global and local installation work as before. Custom applications cannot run with valgrind. default to None (running default executable)

paralleltools.master_mpi(comm, ncores, rank, infileList, RunDir, customAppl)

This is the master for mpi runs taking care of the work distribution and calculating jobs via threading.

Arguments

comminstance of mpi4py.MPI.COMM_WORLD

communicator for the MPI program

ncoresint

number of processes in the MPI-run

rankint

rank of this process (should be 0)

infileListlist

list of of the main files passed to fortran executable as command line argument.

RunDirstr, optional

Specifies path of the fortran-executable (local or global installation). This is available for default-simulation and custom application. (@Developers: Debugging with valgrind always works with a local installation.) None : if executable available within local folder, use local installation, otherwise assume global installation, default ‘’ (empty string) : for global installation ‘./’ : local installation PATH+’/’ : path to executable ending on slash /. Additional steps may be necessary when using this.

customApplstr, optional

define custom executable. Global and local installation work as before. Custom applications cannot run with valgrind. default to None (running default executable)

paralleltools.worker_mpi(comm, rank, infileList, RunDir, customAppl)

This is a worker for mpi runs taking care of calculating jobs received from the master.

Arguments

comminstance of mpi4py.MPI.COMM_WORLD

communicator for the MPI program

ncoresint

number of processes in the MPI-run

rankint

rank of this process (should be 0)

infileListlist

list of of the main files passed to fortran executable as command line argument.

RunDirstr, optional

Specifies path of the fortran-executable (local or global installation). This is available for default-simulation and custom application. (@Developers: Debugging with valgrind always works with a local installation.) None : if executable available within local folder, use local installation, otherwise assume global installation, default ‘’ (empty string) : for global installation ‘./’ : local installation PATH+’/’ : path to executable ending on slash /. Additional steps may be necessary when using this.

customApplstr, optional

define custom executable. Global and local installation work as before. Custom applications cannot run with valgrind. default to None (running default executable)

paralleltools.worker_mpi(comm, rank, infileList, RunDir, customAppl)

This is a worker for mpi runs taking care of calculating jobs received from the master.

Arguments

comminstance of mpi4py.MPI.COMM_WORLD

communicator for the MPI program

ncoresint

number of processes in the MPI-run

rankint

rank of this process (should be 0)

infileListlist

list of of the main files passed to fortran executable as command line argument.

RunDirstr, optional

Specifies path of the fortran-executable (local or global installation). This is available for default-simulation and custom application. (@Developers: Debugging with valgrind always works with a local installation.) None : if executable available within local folder, use local installation, otherwise assume global installation, default ‘’ (empty string) : for global installation ‘./’ : local installation PATH+’/’ : path to executable ending on slash /. Additional steps may be necessary when using this.

customApplstr, optional

define custom executable. Global and local installation work as before. Custom applications cannot run with valgrind. default to None (running default executable)

paralleltools.runMPS_wrapper(infileList, RunDir=None, Debug=False, customAppl=None)

Pass call to tools.runMPS() but handle errors from fortran library. Up to now we ignore those errors and continue with the next job.

For arguments see in tools.runMPS().

paralleltools.runMPS_wrapper(infileList, RunDir=None, Debug=False, customAppl=None)

Pass call to tools.runMPS() but handle errors from fortran library. Up to now we ignore those errors and continue with the next job.

For arguments see in tools.runMPS().