wangjw%usceast@usceast.cs.scarolina.edu (Jingwen Wang) (02/12/91)
The Ease of programming and debugging and efficiency have been the two most critical issues of parallel processing. Both are important, we are unwiiling to sacrafice either one. For some problems, such as computational fluid dynamics, you can never have enough computing power. Efficiency is particularly a difficult thing for massive parallelism. If you are using only a dozen of processors, you might say that ease of programming is most important. So you would prefer a speedup of 4 for easier programming than a speedup of 8 with a difficult programming. However, if you are using a thousand processors and you only receive a speedup of about 30, you would like to sacrifice your ease of programming for better efficiency, provided your problem really needs a substantial speedup (there are such problems that data must be processed in limited time for the results to be meanningful). The importance of programming environment for parallel system is better justified by a recent article in IEEE Computer, Dec. 1990. Wang, Jingwen wangjw@usceast.cs.scarolina.edu ___________________________________ /\ Q /\%\ NASA Means: "Needs Another Seven ________/_\/ \/ Astronauts". /* / |+| The Heart has its reason that /$ / /\ reason does not know. ~~~~ _/ \_ --------------------------------------------------------------------------------
rbe@yrloc.ipsa.reuter.COM (Robert Bernecky) (02/14/91)
The flaw in your reasoning is that you make the assumption that you cannot have programs which are both efficient AND maintainable. If you are writing in stone-age languages such as Fortran, even Fortran 8x,90,9x (pick your value of x), then your assumption is correct: The language is too crude and primitive to meet both needs. What IS needed is a language (or languages) which reflect the way we think: Languages which are concise, which have few rules, and which have simple rules. Once applicative languages such as J have efficient compilers developed for them, they might be able to be the cake you can have and eat. Bob Bernecky Snake Island Research Inc.
eugene@nas.nasa.gov (Eugene N. Miya) (02/21/91)
Call this "computational 'break-even.'" Parallel processing is like controlled thermonuclear fusion......... --e. nobuo miya, NASA Ames Research Center, eugene@orville.nas.nasa.gov {uunet,mailrus,other gateways}!ames!eugene AMERICA: CHANGE IT OR LOSE IT.