So far I've been extremely impressed with Parallel Python. I ran across an oddity I'm not entirely sure how to fix, however.
Using example 1 (sum_primes) on my system (Core 2 Duo iMac) I get the result 3.8 seconds for the parallel version and 6.8 seconds without pp. When I add the following to the top of my code:
import psyco
psyco.full()
The pp version still takes 3.8 seconds, but the non parallel version now takes 0.62 seconds.
So I tried the following:
def sum_primes2(n):
psyco.bind(sum_primes)
return sum_primes(n)
job_server.submit(sum_primes2, (input,), (isprime, sum_primes), ("math", "psyco")))
The parallel version now takes 0.44 seconds
But it looks like I have to build a customized function each time I do this. I can't make a decorator:
def make_psyco(func):
def retval(*args, **kwargs):
psyco.bind(func)
return func(*args, **kwargs)
return retval
sum_primes2 = make_psyco(sum_primes)
Or it tells me that:
Traceback (most recent call last):
File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/pp-1.5.2-py2.5.egg/ppworker.py", line 91, in run
__result = __f(*__args)
File "<string>", line 2, in retval
NameError: global name 'func' is not defined
So I tried:
def func(n):
return sum_primes(n)
With:
job_server.submit(sum_primes2, (input,), (isprime, sum_primes, func), ("math", "psyco")))
That executed just fine, but it is not any more extensible than my first approach.
I modified my decorator with the following two lines:
retval.__name__ = func.__name__
retval.__dict__.update(func.__dict__)
Which worked, but showed no speed improvement over the psycoless version.
Any ideas for how to do this generically?
Forums 
