Newsgroups: comp.parallel
From: "Dennis Gannon" <gannon@cs.indiana.edu>
Subject: Re: parallel c++ class - summary of responses
Organization: Computer Science, Indiana University
Date: Thu, 26 Aug 1993 13:17:58 GMT

Just a follow-up on Russell's note.  First, the good folks at UVA would
probably want to add that Mentat follows a large grain data flow approach
and not a data parallel one.  This is an outstanding and very flexible
model.   Also to echo the "FAQ" document that David Edelsohn has posted,
interested parties should also contact Carl Kesselman at caltech about CC++
which will be out soon. CC++ provides a nice way to extend the C++ type system
to work with thread on a distributed machines (and a bunch of other good ideas
too.)  Jim Larus and his group at wisconsin have defined a nice extension
of C* to C++ called C** which is up and running.  These are just the
language extension gang.

At a very fun meeting hosted by Carl Kesselman and Mani Chandy (the CC++
co-inventor who is well known for lots of other good parallel programming
ideas), the issue of language extensions vs. just using naked C++ libraries
was hotly debated. (A report of this meeting will be circulated soon.)
Ian Angus will do a tutorial on parallelism in C++ at Supercomputing and
I am sure he will have lots to say about this.  sign up!  In addition
there will be a pannel on the subject at SC93.  bring your tomatos. 

Along the library side you should also check out lapack++ when it is out
soon.  (contact roldan pozo pozo@cs.utk.edu).  There are a bunch of other
good parallel C++ library efforts such as those from sandia, llnl and lanl,
but i have not seen public releases of these yet. (am i wrong?)

I have left lots of good stuff out. (see David's FAQ) This is a very exciting
time and good ideas abound.  Check out the proceedings of last year's 
OON-SKI or attend the special session at OOPLSA.  Also look for lots of
other new ideas at next year's OON-SKI (CFP very soon.)

Among the big questions:
what is the right C++ binding for the MPI?  surely the fortran/C version
is not all we want.  The type sytem in C++ can add some help here.
David Walker? any word on this from the committee?

What about I/O systems?  Persistent objects in a MPP context?  (see the UWisc.
work here.)  Anybody using the current generation of MPP systems knows
that this is the real frontier.

When will there be a convergence of ideas?  probably not until we have
had some time to experiment with the ideas out there now.
The vendors of HPC systems will probably also weigh in on this one soon.

sorry for the long posting.
d.
gannon@cs.indiana.edu