Python multiprocessing

From ScientificComputing
Revision as of 07:13, 22 June 2021 by Jarunanp (talk | contribs)

Jump to: navigation, search

< Examples

In this example we show how to launch parallel tasks in Python by using ProcessPoolExecutor in the concurrent.futures module.

"The concurrent.futures module provides a high-level interface for asynchronously executing callables.

 The asynchronous execution can be performed with threads, using ThreadPoolExecutor, or separate processes, using ProcessPoolExecutor. Both implement the same interface, which is defined by the abstract Executor class."


Load modules

Switch to the new software stack

[jarunanp@eu-login-05 ~]$ env2lmod

or, set your default software stack to the new software stack

[jarunanp@eu-login-05 ~]$ new

Load a Python module

[jarunanp@eu-login-05 ~]$ module load gcc/6.3.0 python/3.8.5


Create a project folder

[jarunanp@eu-login-05 ~]$ mkdir python_multiprocessing
[jarunanp@eu-login-05 ~]$ cd python_multiprocessing
[jarunanp@eu-login-05 python_multiprocessing]$

Open a new file named with a text editor and add the following code:

from concurrent.futures import ProcessPoolExecutor
import sys
import time
import numpy as np

def accumulate_sum(v):
    sumv = 0
    for i in v:
        sumv += i
    return sumv

def main(): 
    n = 50_000_000
    vec = np.random.randint(0,1000,n)
    # The script requires an input argument which is the number of processes to execute the program
    num_processes = int(sys.argv[1])
    n_per_process = int(n/num_processes) 
    vec_per_process = [vec[i*n_per_process:(i+1)*n_per_process] for i in range(num_processes)]
    # start the stop watch
    start = time.time()

    with ProcessPoolExecutor(max_workers=num_processes) as executor:, vec_per_process)

    # end the stop watch
    end = time.time()

    print("The accumulated sum is {:3.2e}".format(sum(results)))
    print("Elasped time: {:3.2f}s".format(end-start))
if __name__ == '__main__':

Request an interactive session on a compute node

[jarunanp@eu-login-05 python_multiprocessing]$ bsub -n 5 -Is bash
Generic job.
Job <176062929> is submitted to queue <normal.4h>.
<<Waiting for dispatch ...>>
<<Starting on eu-c7-112-13>>
FILE: /sys/fs/cgroup/cpuset/lsf/euler/job.176062929.12449.1624344257/tasks
[jarunanp@eu-c7-112-13 python_multiprocessing]$ 

Launch the Python script with num_processes = 1, 2 and 4

[jarunanp@eu-c7-112-13 python_multiprocessing]$ python 1
The accumulated sum is 2.50e+10
Elasped time: 14.10
[jarunanp@eu-c7-112-13 python_multiprocessing]$ python 2
The accumulated sum is 2.50e+10
Elasped time: 7.88
[jarunanp@eu-c7-112-13 python_multiprocessing]$ python 4
The accumulated sum is 2.50e+10
Elasped time: 4.75

From the output, increasing number of processes reduced the runtime to execute the operations. The speedup was around 2 and 3 times for num_processes = 2 and 4, respectively. It is not linear but we could gain a significant factor of runtime.

< Examples