100 threads that append a random number to a list 20000 times

append2.py

import thread
import time
import random

big_list = []
thread_status = {}

def appendo(n):
    thread_status[thread.get_ident()] = "running"
    for x in xrange(n):
        big_list.append(random.randint(1,5000))
    thread_status[thread.get_ident()] = "stopped"

tl = [ thread.start_new_thread(appendo,(20000,)) for x in range(100) ]
print len(tl)

time.sleep(0)
while 1:
    if 'running' in thread_status.values():
        time.sleep(0)
    else:
        break

print "the size of big_list", len(big_list)
print "sum: ", sum(big_list)

Thread Overhead

I ran the above script for two invocations, one with 100 threads writing 20k random ints to a list the other with 1 thread writing 2M random ints. So that same amount of work, one thread verses 100 threads.

cpython

$ time python append_list2.py 
100
the size of big_list 2000000
sum:  4995798476

real    0m6.199s
user    0m5.706s
sys 0m2.238s

$ time python append_list2.py 
1
the size of big_list 2000000
sum:  4998149729

real    0m5.335s
user    0m4.863s
sys 0m1.866s

pypy

$ time pypy append_list2.py 
100
the size of big_list 2000000
sum:  5003485315

real    0m0.328s
user    0m0.230s
sys 0m0.098s

$ time pypy append_list2.py 
1
the size of big_list 2000000
sum:  5002606777

real    0m0.291s
user    0m0.203s
sys 0m0.088s