massively parallel smalltalk
Jecel Assumpcao Jr
jecel at merlintec.com
Sat Apr 28 00:55:37 UTC 2001
While I really like Linda (and have seen at least two nice
implementations in Smalltalk), it does require using a very different
design for your programs.
Several other alternatives have been developed for massively parallel
Smalltalk and I would recomend at least a quick look at:
http://cva.stanford.edu/j-machine/cva_j_machine.html
Though this Concurrent Smalltalk (there was another project with the
same name) had a Lisp-like syntax instead of the keywords we love so
much, it really was Smalltalk and was running on 1024 node machines.
Hmmm... I see only a brief mention of "distributed objects" in there,
though I was pretty sure that was where the notion of "concurrent
aggregates" came from. The idea was that if collections were allocated
with its members spread out through several nodes, messages such as
#do:, #select: and so on would automatically express parallel
execution. So many "normal" Smalltalk programs could become highly
parallel simply by using concurrent versions of collection objects
instead of having to rewrite everything to use tuples.
-- Jecel
More information about the Squeak-dev
mailing list
|