Would you know which of these implementations are the most mature or
the most used? Stability would be better than fancy features.
Thanks,
Edward
Conrad Hinsen seems to still be actively developing Scientific Python, and it's MPI implementation has been around long enough that it should be pretty mature. So, I suspect that there would be reasonable support for bugfixes if needed.
The other two that appear to be actively developed (and in use in real
world projects) are MMPI (
http://www.penzilla.net/mmpi/index.shtml) and
MyMPI (
http://peloton.sdsc.edu/~tkaiser/mympi/).
Looking more closely at the issue of 'can pass arbitrary python objects' vs. 'can only pass Numeric arrays and basic types like strings', I realise now that it isn't really a big deal, since anything that can be pickled (ie "serialized") can be sent as a string and then unpickled ... and my guess is this how most python MPI implementations that support 'object' passing do it. Note that there are limitations on pickling classes (
http://docs.python.org/lib/node66.html), but class *instances* retain their attributes and so should be passable.
Andrew