[ale] Re: Threading? (was: Re: [ale] C#)

[Michael D. Hirsch]

Thanks for setting me straight, Joe.  I'm glad to hear that i was 
overstating things.  I do know that we had tremendous performance 
problems with ZOPE which I was told was attributed to threading (more 
likely GIL) issues.  Now I don't know what caused them.

Hmm.

Michael

On Thursday 26 September 2002 05:27 pm, you wrote:
> Michael D. Hirsch wrote:
>
>
> [A lot of good stuff about threading in general...]
>
> > Python only has user threads (I think--I'm no expert on this aspect
> > of python).  When a thread does I/O the threading library is pretty
> > clever and can time it out, but until it either times out or
> > finishes, no other thread in that process will proceed.  So in,
> > say, a web server like ZOPE, if there are a lot of connections
> > there will be a lot of threads blocking on I/O.  This will slow it
> > down incredibly.  Also if the threads have to pull data off the
> > disk they will block on disk access, again slowing it down.  For
> > this reason ZOPE tries to cache all it's objects in RAM.  It is
> > quite possible to run out of RAM for this. Also, until everything
> > is in RAM there are lots of threads blocking on disk access.
> >
> > I think I read somewhere that python is trying to use kernel
> > threads, but I don't recall if that was in version 2.2, which is
> > already release, or in a future version.
>
> Python does use kernel threads, and has for some time (possibly
> since threads were originally implemented in Python, but I'm not
> sure about that).
>
> I/O operations don't block other threads in Python. There
> *is* a deficiency (some would say) in CPython threads, in that
> the Python virtual machine executes Python bytecode under
> the Global Interpreter Lock. That means that only one thread
> can execute Python bytecode at a time. However, native code
> running under Python, including the I/O facilities and most
> extension modules, releases the GIL so that other threads
> can proceed. So Zope doesn't suffer any performance loss
> due to the GIL. But it is the case that multiple threads
> running Python bytecode can't take advantage of multiple
> CPUs, which sucks IMO.
>
> Note also that Jython (Python compiled to Java bytecode) uses
> the Java threading facilities and does not have anything
> like the GIL, so a Jython solution should be as fast as
> a Java solution, and the code will be *way* prettier :-)
>
> > I love python--I would say it is my favorite language--and I hope
> > it gets better threading support soon.
>
> I think Guido is not much interested in losing the GIL.
> The Stackless guys might, though.
>
> Cheers,
>
> -- Joe


---
This message has been sent through the ALE general discussion list.
See http://www.ale.org/mailing-lists.shtml for more info. Problems should be
sent to listmaster at ale dot org.