GIL cpu multi core usage problem

Pau Freixes pfreixes at gmail.com
Fri Jun 13 13:55:30 CEST 2008


Hi,

When you say this "C extensions (usually) release the GIL when they don't
call into any Python code" do you talk about this macros ?

Py_BEGIN_ALLOW_THREADS

Py_END_ALLOW_THREADS



On Wed, Jun 11, 2008 at 3:52 AM, Gabriel Genellina <gagsl-py2 at yahoo.com.ar>
wrote:

> En Mon, 09 Jun 2008 15:26:09 -0300, Pau Freixes <pfreixes at gmail.com>
> escribió:
>
>  Surly this is a recurring theme into python dev world, but I need your
>> help
>> for confirm if the follow image it's really
>>
>> http://www.milnou.net/~pfreixes/img/cpu_usage_gil_problem.png<http://www.milnou.net/%7Epfreixes/img/cpu_usage_gil_problem.png>
>> <http://www.milnou.net/%7Epfreixes/img/cpu_usage_gil_problem.png>
>>
>> I'm writing a brief article for my blog and I need to make sure about the
>> current problem with GIL and multi core environments, this picture try to
>> explain with images the problem for scheduling multiple threads running
>> python code of same interpreter into multiple cpu cores. Can  anyone
>> confirm
>> to me this picture ?
>>
>
> Yes, if both threads are executing pure Python code, they can't run
> simultaneously.
>
> Note that:
> - C extensions (usually) release the GIL when they don't call into any
> Python code
> - The interpreter also releases the GIL before any I/O operation
> If the process is I/O bound then the GIL is not so important, and if it's
> CPU bound you may benefit from reimplementing the critical parts in C (by
> example, using NumPy for a number-crunching process). Another alternative is
> to use multiple processes with some form of IPC instead of multiple threads.
> These are the usual arguments against "fear of GIL". You should consider
> your specific application to see if the GIL is actually a problem in your
> case or not.
>
> --
> Gabriel Genellina
>
> --
> http://mail.python.org/mailman/listinfo/python-list
>



-- 
Pau Freixes
Linux GNU/User
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-list/attachments/20080613/86fe2e49/attachment.html>


More information about the Python-list mailing list