Have a Question?
Parallelization using OpenMP
Why OpenMP?
Open Multi Processing is designed for parallel programming on shared-memory architectures, where multiple processors or cores share the same address space. Threads within a process can communicate and synchronize through shared variables. OpenMP is designed for parallel programming on shared-memory architectures, where multiple processors or cores share the same address space. In this article we’ll use the python standard multiprocessing library to run a task/process for x times in a single compute node.
Step 1: Create the Python Script
$ cd /path/to/your/directory
Use a text editor of your choice to create a Python script named hello_openmp.py. This script will print “Hello, World!” but in parallel using OpenMP.
import os
#import the python standard multiprocessing library
import multiprocessing as mp
def print_hello(_):
print(f"Hello ARCC2 | Process ID: {os.getpid()}")
if __name__ =="__main__":
# Set the number of tasks or precesses:
# This is as same as --ntasks-per-node
num_tasks = 10
# use mp.pool and run print_hello function:
with mp.Pool(processes=num_tasks) as pool:
pool.map(print_hello, range(num_tasks))
Step 2: Create the Slurm Job Script
Create a Slurm job script named hello_openMP.sh using a text editor:
#!/bin/bash
#SBATCH --job-name=hello_openMP
#SBATCH --output=hello_openMP.out
#SBATCH --error=hello_openMP.err
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=10
#SBATCH --cpus-per-task=1
#SBATCH --time=00:01:00
#SBATCH --qos=proj-xxx
# Load the necessary modules:
# Run the Python Script with OpenMP parallelization:
python3.6 hello_openMP.py
Now, let’s go through each part of the script:
-
#!/bin/bash:- Purpose: This line is called the shebang, and it specifies the shell to be used to interpret the script. In this case, it’s Bash.
-
#SBATCH --job-name=openmp_hello:- Purpose: Specifies the name of the job. It will be used in job listings and output files.
-
#SBATCH --output=openmp_hello.out:- Purpose: Specifies the file to which the standard output (stdout) of the job will be written.
-
#SBATCH --error=openmp_hello.err:- Purpose: Specifies the file to which the standard error (stderr) output of the job will be written.
-
#SBATCH --nodes=1:- Purpose: Requests a total number of nodes for the job. 1 in this example.
-
#SBATCH --ntasks-per-node=10:- Purpose: Specifies the number of tasks (processes) to be launched on each node. 10 in this example.
-
#SBATCH --cpus-per-task=1:- Purpose: Specifies the number of CPU cores assigned to each task. This means each hello world function in our python script is task and will use 1 CPU cores.
-
#SBATCH --time=00:01:00:- Purpose: Specifies the maximum time the job is allowed to run. 1 minute(s) (00:01:00) in this example.
-
#SBATCH --qos=proj-xxx:- Purpose: Specify the quality of service allocated.
-
python3.6 hello_openMP.py:- Purpose: Executes the Python script
hello_openMP.pythat contains the “Hello World” program with OpenMP parallelization.
- Purpose: Executes the Python script
Up Next: Parallelization using MPI (Message Passing Interface)
In the next article parallelization of a simple task across two nodes with distributed memory using Message Passing Library is discussed.