Parallelization parameters

You can set parameters to customize parallel execution in Abaqus.

Parallelization in Abaqus is discussed in detail in the following sections:

The relevant parameters are listed here.

auto_convert

If this parameter is set equal to ON and an Abaqus/Explicit analysis is run in parallel with parallel=domain, the convert=select, convert=state, and convert=odb options will be run automatically at the end of the analysis if required. The default value is ON.

cpus

Number of processors to use during an analysis run if parallel processing is available. The default value for this parameter is 1.

domains

The number of parallel domains in Abaqus/Explicit. If the value is greater than 1, the domain decomposition will be performed regardless of the values of the parallel and cpus variables. However, if parallel=DOMAIN, the value of cpus must be evenly divisible into the value of domains. If this parameter is not set, the number of domains defaults to the number of processors used during the analysis run if parallel=DOMAIN or to 1 if parallel=LOOP.

gpus

Activate direct solver acceleration using GPGPU hardware in Abaqus/Standard. The value of this parameter should be the number of GPGPUs to use for an analysis. In an MPI-based parallel Abaqus/Standard analysis, this is the number of GPGPUs to use on each host.

max_cpus

Maximum number of processors allowed if parallel processing is available. If this parameter is not set, the number of processors allowed equals the number of available processors on the system.

mp_file_system

Type of file system available for an MPI-based parallel Abaqus analysis. The parameter must be set to a tuple; for example,

mp_file_system=(SHARED,LOCAL)

The first item in the tuple refers to the directory where the job was submitted, while the second refers to the job's scratch directory. If the file system hosting a directory is LOCAL, Abaqus will copy the required analysis files to the remote host machines and, at the end of the run, copy the output files back. In this case it must be possible to create the job's directory structure on all the hosts in mp_host_list. A SHARED file system means that the host machines share the same file system and file transfer is not necessary. With the recommended default (DETECT, DETECT) setting, Abaqus will determine the type of file system that exists. An MPI-based parallel Abaqus/Explicit analysis will use the scratch directory only if a user subroutine is used, whereas Abaqus/Standard normally writes large temporary files in this directory. Running on a local file system will generally improve the performance.

mp_host_list

List of host machine names to be used for an MPI-based parallel Abaqus analysis, including the number of processors to be used on each machine; for example,

mp_host_list=[['maple',1],['pine',1],['oak',2]]

indicates that, if the number of cpus specified for the analysis is 4, the analysis will use one processor on a machine called maple, one processor on a machine called pine, and two processors on a machine called oak. The total number of processors defined in the host list has to be greater than or equal to the number of cpus specified for the analysis. If the host list is not defined, Abaqus will run on the local system. When using a supported queuing system, this parameter does not need to be defined. If it is defined, it will get overridden by the queuing environment.

mp_mode

Set this variable equal to MPI to indicate that the MPI components are available on the system. Set mp_mode=THREADS to use the thread-based parallelization method. The default value is MPI where applicable.

mp_mpi_implementation

This variable determines the underlying MPI implementation to use. Generally, this variable does not need to be specified.

mp_mpirun_options

String of options that are passed to the MPI launcher for an MPI-based parallel Abaqus analysis. Generally this variable does not need to be specified.

mp_mpirun_path

A dictionary to define the full path to the MPI launcher for a given MPI implementation. For example, on Windows,

mp_mpirun_path={NATIVE: 'C:\\Program Files\\
   Microsoft HPC Pack 2008 R2\\bin\\mpiexec.exe'}
mp_num_parallel_ftps

When performing parallel file staging using MPI-based parallelization, this parameter controls the number of simultaneous MPI file transfers. The first item controls the transfer of files to and from the temporary scratch directory. The second item controls the transfer of files to and from the analysis working directory. Setting either value to 1 disables the parallel file staging process. The use of file staging depends on the values specified in mp_file_system.

mp_rsh_command

Preferred command to open a remote shell on the machines specified by mp_host_list. Abaqus needs to open a remote shell to create and remove directories and files if the file system is not shared. The default value for this option is platform dependent; for example,

mp_rsh_command='ssh -n -l %U %H %C'

The following placemarkers are used:

%U

Username.

%H

The host where the remote shell is opened.

%C

The command to be executed on the host.

Abaqus automatically uses secure copy (scp) to copy files to and from remote hosts if this parameter is set to use secure shell. By default, this parameter is ignored in favor of built-in MPI rsh/scp commands.

order_parallel

The default direct solver ordering mode in Abaqus/Standard if you do not specify the parallel ordering mode on the abaqus command line. If this parameter is set equal to OFF, the solver ordering will not be performed in parallel. If this parameter is set equal to ON, the solver ordering will be run in parallel. The default for parallel solver ordering is ON.

parallel

The default parallel equation solution method in Abaqus/Explicit if the user does not specify the parallel method on the abaqus command line. Possible values are DOMAIN or LOOP; the default value is DOMAIN.

standard_parallel

The default parallel execution mode in Abaqus/Standard if you do not specify the parallel mode on the abaqus command line. If this parameter is set equal to ALL, both the element operations and the solver will run in parallel. If this parameter is set equal to SOLVER, only the solver will run in parallel. The default parallel execution mode is ALL.