Which multiprocessing methods use shared memory?

[John Nagle]

On 7/27/2010 12:30 PM, MRAB wrote:
> Kevin Ar18 wrote:
>> I'm not sure my previous message went through (I wasn't subscribe), so
>> I'm gonna try again.
>>
>> The multiprocessing module has 4 methods for sharing data between
>> processes:
>> Queues
>> Pipes
>> Shared Memory Map
>> Server Process
>>
>> Which of these use shared memory?
>>
>> I understand that the 3rd (Shared Memory Map) does, but what about
>> Queues?
>>
> The documentation says:
>
> """class multiprocessing.Queue([maxsize])
> Returns a process shared queue implemented using a pipe and a few
> locks/semaphores. When a process first puts an item on the queue a
> feeder thread is started which transfers objects from a buffer into the
> pipe."""

    Python's "shared memory" between processes is un-Pythonic.  You
can't share Python object, only C objects.  The locking mechanisms
are very limited, and slow; locking actually takes place via
messages over pipes.  There's no dynamic allocation in the shared area.

    Realistically, if you're using Python, you're not that concerned
about compute speed.  So don't bother with shared memory, which is
a performance optimization.

					John Nagle