multiprocessing pystone benchmark

" Simple multiprocessing test.pystones benchmark "
" Anders Wallin 2008Jun15 anders.e.e.wallin (at) gmail.com "
from test import pystone
import processing
import time

STONES_PER_PROCESS= 10*pystone.LOOPS

def f(q):
    t=pystone.pystones(STONES_PER_PROCESS)
    q.put(t,block=True)

if __name__ == '__main__':
    print 'multiprocessing test.pystones() benchmark'
    print 'You have '+str(processing.cpuCount()) + ' CPU(s)'
    print 'Processes\tPystones\tWall time\tpystones/s'

    results = processing.Queue()
    for N in range(1,processing.cpuCount()+3):
        p=[]
        q=processing.Queue()
        results=[]

        for m in range(1,N+1):
            p.append( processing.Process(target=f,args=(q,)) )

        start=time.time()
        for pr in p:
            pr.start()
        for r in p:
            results.append( q.get() )
        stop=time.time()

        cputime = stop-start

        print str(N)+'\t\t'+str(N*STONES_PER_PROCESS) \
              +'\t\t'+ str(cputime)+'\t'+str( N*STONES_PER_PROCESS / cputime )

5 thoughts on “multiprocessing pystone benchmark”

1. is it better now?

2. get rid of the backquotes `` and use if __name__ = '__main' if you'd like others to be able to run this without errors and post their results.

3. '''Here is a slightly improved version that actually runs

   Simple multiprocessing pystone.pystones() benchmark '''
   ''' Anders Wallin 2008Jun15 anders.e.e.wallin (at) gmail.com '''
   from test import pystone
   import processing
   import time

   STONES_PER_PROCESS= 10*pystone.LOOPS

   def f(q):
   t=pystone.pystones(STONES_PER_PROCESS)
   q.put(t,block=True)

   if __name__ == '__main__':
   processing.freezeSupport()

   print 'multiprocessing test.pystones() benchmarkn'
   print 'You have '+str(processing.cpuCount()) + ' CPU(s)n'
   print 'ProcessestPystonestWall timetpystones/stSpeedup'

   results=processing.Queue()
   for N in range(1,processing.cpuCount()+3):
   p=[]
   q=processing.Queue()
   results=[]

   for m in range(1,N+1):
   p.append( processing.Process(target=f,args=(q,)) )

   start=time.time()
   for pr in p:
   pr.start()
   for r in p:
   results.append( q.get() )
   stop=time.time()

   cputime = stop-start
   if N==1: singlecpurate = N*STONES_PER_PROCESS / cputime

   print "%utt%utt%.1ftt%.ftt%.1fX" % (N, N*STONES_PER_PROCESS,
   cputime, N*STONES_PER_PROCESS / cputime,
   N*STONES_PER_PROCESS / (cputime * singlecpurate) )

4. Thanks for the update, I spent some time changing it myself too. But my indentation got screwed up.
   You wanted 8 cores, here it is 🙂

   Xeon DP 2.66ghzx2 (E5430) = 8 cores total

   multiprocessing test.pystones() benchmark

   You have 8 CPU(s)

   Processes Pystones Wall time pystones/s Speedup
   1 500000 6.9 72727 1.0X
   2 1000000 6.8 147449 2.0X
   3 1500000 6.8 221729 3.0X
   4 2000000 6.8 292912 4.0X
   5 2500000 6.8 365283 5.0X
   6 3000000 6.8 438340 6.0X
   7 3500000 6.9 506806 7.0X
   8 4000000 6.9 577868 7.9X
   9 4500000 10.2 441739 6.1X
   10 5000000 11.3 441384 6.1X

5. There is something going on with the quotes... I get a 'non ASCII-character' error with your program, but it runs OK if I change all the single-quotes to double-quotes.
   Perhaps you're on a unicode-only system?

   Thanks for the results, I've added them to the graph.

   As a further exercise I'm thinking of comparing pystones or another benchmark using different speedup techniques, i.e. psyco, cython etc,
   shedskin looks very interesting: http://code.google.com/p/shedskin/

Comments are closed.