parallel_write.py

import os
import threading

NUM_THREADS = 8
FILE_SIZE = 4 * 1024 * 1024 * 1024
BLOCK_SIZE = 1024 * 1024
BLOCK = b'\0' * BLOCK_SIZE

def worker(fd, offset, length):
    for i in range(offset, offset + length, BLOCK_SIZE):
        os.pwrite(fd, BLOCK, i)

fd = os.open("foo.dat",  os.O_CREAT | os.O_TRUNC | os.O_WRONLY)
try:
    threads = []
    for i in range(NUM_THREADS):
        size = FILE_SIZE // NUM_THREADS
        thread = threading.Thread(target = worker, args = (fd, i * size, size))
        threads.append(thread)
        thread.start()

    for thread in threads:
        thread.join()
    os.fsync(fd)
finally:
    os.close(fd)