On 14 January 2013 02:46, Mark Janssen <[email protected]> wrote: > On Sun, Jan 13, 2013 at 8:37 PM, Oscar Benjamin > <[email protected]> wrote: >> On 14 January 2013 02:33, Mark Janssen <[email protected]> wrote: >>> Lol, well that's why I'm asking. I don't see how they can do it >>> without considerable difficulties. >> >> What do you want the GIL for across machines? The normal purpose of >> the GIL is to preserve the integrity of Python's in-memory data >> structures. These are only accessible within one process so what good >> would a multi-process GIL be? > > Excuse me, I actually thought you knew what you were talking about.
Steady on... > A Beowulf cluster, by it's nature, is across many-machines. Now you > can't have a GIL and a single python program across many-machines > without some tricks. I'm just wondering how (or if) they solved that > problem.... You've used the word 'program'. I used the word 'process'. I have used HPC clusters that I assume count as Beowulf clusters. However, I have never used them in such a way that a single process occurred on multiple machines (the idea seems absurd to me). Maybe I've missed something here... Oscar -- http://mail.python.org/mailman/listinfo/python-list
