线程模块与多处理模块的比较

Question

因此，我试图比较线程处理是更快还是多处理。从理论上讲，由于GIL，多处理应该比多线程更快，因为一次只运行一个线程。但是我得到了相反的结果，即线程比多处理花费的时间少，我缺少什么请帮助

下面是线程的代码

import threading
from queue import Queue
import time

print_lock = threading.Lock()

def exampleJob(worker):
    time.sleep(10)  
    with print_lock:
        print(threading.current_thread().name,worker)


def threader():
    while True:

        worker = q.get()


        exampleJob(worker)


        q.task_done()

q = Queue()

for x in range(4):
     t = threading.Thread(target=threader)

     print(x)
     t.daemon = True


     t.start()

start = time.time()


for worker in range(8):
    q.put(worker)


q.join()


print('Entire job took:',time.time() - start)

以下是多处理的代码

import multiprocessing as mp
import time

def exampleJob(print_lock,worker):                 # function simulating some computation
    time.sleep(10)
    with print_lock:
        print(mp.current_process().name,worker)

def processor(print_lock,q):                       # function where process pick up the job
    while True:
        worker = q.get()
        if worker is None: # flag to exit the process
            break
        exampleJob(print_lock,worker)


if __name__ == '__main__':

    print_lock = mp.Lock()
    q = mp.Queue()
    processes = [mp.Process(target=processor,args=(print_lock,q)) for _ in range(4)]

    for process in processes:
        process.start()    

    start = time.time()
    for worker in range(8):
        q.put(worker)

    for process in processes:
        q.put(None) # quit indicator

    for process in processes:
        process.join()

    print('Entire job took:',time.time() - start)

Answer 1

这不是一个合适的测试time.sleep可能没有获得GIL，因此您运行的是并发线程而不是并发进程。线程速度更快，因为没有启动成本

您应该在线程中执行一些计算，然后您将看到差异

Answer 2

由于存在GIL，只有在执行计算密集型任务时，添加到@zmbq线程的速度才会较慢。如果您的操作是I/O绑定的，并且很少有其他类似的操作，那么线程肯定会更快，因为所涉及的开销更少。请参阅下面的博客，以更好地了解相同的内容

Exploiting Multiprocessing and Multithreading in Python as a Data Scientist

希望这有帮助