Python multi-thread ,multiprocessing, GIL

In CPython, the global interpreter lock, or GIL, is a mutex that protects access to Python objects, preventing multiple threads from executing Python bytecodes at once. The GIL prevents race conditions and ensures thread safety.
In short ,Python global intepreter lock will try prevent multiple thread from executing at same time to ensure thread safe . Lets see how it impact multi-threaded processing .

The test

Lets test on both windows by exeucting simple cpu-bound and io-bound task .
cpu task — check prime number from 1–20000 .
io task — generate 20k guid write into csv file1000 rows per file . clean up all files after finish

test execution time for single threaded , multi-threaded and multi-processing

Implementation

import csv 
import uuid
import threading
import time
import multiprocessing
import os
def io_task(start, end):
with open(f'{start}_{end}.csv', mode='w') as csv_file:
for i in range(start, end):
writer = csv.writer(csv_file, delimiter=',', quotechar='"', quoting=csv.QUOTE_MINIMAL)
writer.writerow(['value'])
writer.writerow([uuid.uuid4()])
os.remove(f'{start}_{end}.csv')def is_prime(n):
for i in range(2,n):
if n % 2 == 0:
return False
return True
def cpu_task(arr):
for i in arr:
is_prime(i)
def single_thread_cpu():
start = time.time()
total = 20000
cpu_task([x for x in range(2,total)])
end = time.time()
print(f'[single thread cpu task ] total : {(end-start)} sec')
def single_thread_io():
s = time.time()
total = 500000
io_task(0, total)
e = time.time()
print(f'[single thread io task ] total : {(e-s)} sec')
def proc_do_cpu_task():
start = time.time()
total = 20000
proc_count = 5
pool = multiprocessing.Pool(proc_count)
parr = []
for i in range(2, proc_count + 1):
nums = [x for x in range(2,total) if x%i == 0]
p = pool.apply_async(cpu_task, args=(nums,))
parr.append(p)
for p in parr:
p.get()
pool.close()
pool.join()
end = time.time()
print(f'[process cpu task ] total : {(end-start)} sec')
def thread_do_cpu_task():
start = time.time()
total = 20000
thread_count = 5
tarr = []
for i in range(2, thread_count+1):
nums = [x for x in range(2,total) if x%i == 0]
t = threading.Thread(target=cpu_task, args=(nums,))
tarr.append(t)
t.start()
for t in tarr:
t.join()
end = time.time()
print(f'[thread cpu task ] total : {(end-start)} sec')
def proc_do_io_task():
s = time.time()
total = 500000
proc_count = 5
pool = multiprocessing.Pool(proc_count)
parr = []
for i in range(0, total, 1000):
start, end = i, i+1000
p = pool.apply_async(io_task, args=(start, end,))
parr.append(p)
for p in parr:
p.get()
pool.close()
pool.join()
e = time.time()
print(f'[process io task ] total : {(e-s)} sec')
def thread_do_io_task():
s = time.time()
total = 500000
tarr = []
for i in range(0, total, 1000):
start, end = i, i+1000
t = threading.Thread(target=io_task, args=(start, end,))
tarr.append(t)
t.start()
for t in tarr:
t.join()
e = time.time()
print(f'[thread io task ] total : {(e-s)} sec')
if __name__ == "__main__":
single_thread_cpu()
#single_thread_io()
#thread_do_cpu_task()
#proc_do_cpu_task()
#thread_do_io_task()
#proc_do_io_task()

The Results

Windows (4 Core i5)
[single thread cpu task ] total : 14.984469652175903 sec
[thread cpu task ] total : 9.49857497215271 sec
[process cpu task ] total : 6.357367992401123 sec
[single thread io task ] total : 9.06583833694458 sec
[thread io task ] total : 10.542978048324585 sec
[process io task ] total : 7.662624359130859 sec

Ubuntu 18.04 vm (2 Cores)
[single thread cpu task ] total : 9.855742931365967 sec
[thread cpu task ] total : 4.477677345275879 sec
[process cpu task ] total : 2.953005313873291 sec

[single thread io task ] total : 5.596938371658325 sec
[thread io task ] total : 17.54924726486206 sec
[process io task ] total : 3.260897636413574 sec

Conclusion

In multiprocessing GIL is independent . so always try to go multiprocessing when possible if for speeding up. it is faster and processing logic fully decoupled ,when sub-process died will not kill whole process .

always test the result before going multi-threaded for both io bound and cpu bound task , some times multi-threaded even slower than single threaded .

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store