Newsgroups: comp.parallel.mpi,comp.parallel.pvm From: Kamran Karimi Subject: Re: Why explicit message passing?? Organization: World Lynx, Inc. Date: Thu, 2 Apr 1998 03:04:54 +0000 Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Message-ID: Bruce Scott TOK wrote: >I think the biggest reason we don't see number crunchers using OO type >code is that compilers aren't good enough. See Eugene Miya's FAQ pieces >about people asking for "automatic parallelism" since that is what the >OO advocates appear to want. The issues of running a program in a parallel fashion or a distributed fashion are a bit different. A distributed program can be a parallel program, though. My main argument is about distributed programming. The most important point here is that the same programming model be used in non-distributed and distributed programming. This will result in more people being able to write distributed programs, or better yet, the same program being usable in both distributed and non-distributed environments. In OOP you can hide the location of an object, allowing the programmer to use an object's services without bothering about its location. This is very good. >In every case I've heard of, the automatic distribution systems appear >to do OK on a few PEs but on many PEs they don't see the speedup one can >easily get with things like MPI. > >Until this changes, explicit message passing will be the rule. Hey, >some of us think that way natively, anyway! I wish they would allocate more research resources to such issues. As an incentive for this, I think we should show some more interest in these alternative solutions. -Kamran