Path: utzoo!attcan!uunet!cs.utexas.edu!usc!henry.jpl.nasa.gov!elroy.jpl.nasa.gov!zardoz!dhw68k!stein From: stein@dhw68k.cts.com (Rick 'Transputer' Stein) Newsgroups: comp.software-eng Subject: Re: (none) Message-ID: <27779@dhw68k.cts.com> Date: 24 Nov 89 17:31:14 GMT References: <8911192227.AA12944@cs2.cs.wsu.edu> Reply-To: stein@dhw68k.cts.com (Rick 'Transputer' Stein) Organization: Wolfskill & Dowling residence; Anaheim, CA (USA) Lines: 182 In article <8911192227.AA12944@cs2.cs.wsu.edu> dbenson@CS2.CS.WSU.EDU (David B. Benson) writes: >Subject: Congress Finds Bugs in the Software > > [AAAS Science, 10 November 1989, vol. 246, p. 753] > [by M. Mitchell Waldrop] [Introduction to the essay deleted by me, cut to the chase...] > > "Software," says the report, "is now the choke point in > large systems." I've include an essay I've been trying to get published in the New York Times, the Wall Street Journal, and now the Washington Post (still haven't heard from the Post yet). I'm very glad to see that some members of the Congress and the leadership in this country are beginning to wake up. Supercomputerphobia by Richard M. Stein The Bush Administration has declared the Japanese unfair traders for their reluctance to purchase U.S. made supercomputer products. This proclamation not only intensifies the alienation between our nations, but is an erroneous and misguided attempt to preserve an industry which is in decline. Despite popular claims, supercomputer hardware fulfills a diminishing portion in the technological equation of invention and economic self-sufficiency. In a knee-jerk reaction to the perceived shortfall in supercomputing preeminence, The Office of Science and Technology Policy presented "The Federal High Performance Computing Program" to the Congress on September 8. It proposes $1.9 billion in spending over several years for developing prototype supercomputers, a high-speed optical fiber computer network linking the nation's supercomputer centers, funds to develop operating systems and support software, and investment in university education to produce 1000 computer science Ph.D.s a year by 1995. While laudable in scope, the Initiative is flawed. It does not address the development of application software, the most valuable part of a computer system. Application software makes supercomputers the indispensable engines of science and industry. Without software, any computer, let alone a supercomputer, is incapable of doing any work. Our nation is highly dependent on computer simulation to predict and posit solutions in virtually every sector of the economy and society. Automobiles, aircraft, powerplants, pharmaceutical products, pollution controls, and the weather are modelled with computer software simulations. Application software is a strategic commodity. The FSX agreement with Japan excluded the delivery of the operational flight plan software, the essential computer programs which control the fighter aircraft's guidance, flight, and weapons system computers. ICBMs achieve their accuracy from on-board guidance computers driven with inertial navigation software. Computer simulation, the substitution of mechanized and electronic agents for human activity, has given us ubiquitous freedom and convenience. No longer must we wait on established banking hours to get cash, the automatic teller machine provides 24 hour access. The electronic banking revolution of the 80's happened because some software engineers figured out how to imitate a bank teller through computer simulation. Arbitragers conduct program trading by monitoring stocks. Arbitrage is made possible by simulating the simultaneous actions of many clerks comparing stock and futures data. Bigger computers examine more hedges faster, giving an individual brokerage house a slight advantage over a lesser equipped firm, and a huge advantage over the small investor. Conducting trades a few milliseconds faster than a competitor, when large volumes are at stake, gives the faster, more powerful trader the advantage. Time is money. The extraordinary safety records of jet aircraft arise from automated design, engineering, and the computerized testing of structures, engines, and controls. Mechanical computer aided engineering (MCAE) has made possible the analysis of many configurations and assemblages of parts. With MCAE, a small group of engineers expeditiously decide which configuration provides a greater safety margin because they can simulate vibrations, stresses, and pressures automatically with a computer simulation. This testing activity used to require many man-hours, if not years to accomplish with manual techniques, and now routinely occurs in minutes. Yet, the rate of increase in computer speed is slowing because advances in semiconductor manufacturing are reaching theoretical limits. This implies that existing computer simulations, and the software written to execute them, will no longer speed up. As far as supercomputers are concerned, speed improvements will be more difficult to achieve. This means that we can expect smaller and fewer gains in personal productivity. This fact has broad implications for our entire society. The growing need to solve problems will force us to investigate phenomena through computer simulations. And as the simulations become more complex, we must apply ever greater computational resources to timely resolve and derive solutions. How can we perform more complex and articulated simulations if the existing computing bandwidth has an asymptotic throughput? The answer is simple: parallel processing. Suppose that you want to calculate 100 spreadsheets, but have only 1 processor, like a personal computer. Then you must compute each spreadsheet sequentially on that processor. If however you had 100 processors, then you could calculate all 100 spreadsheets simultaneously by placing each on a separate processor. This form of computation is what parallel processors are about: processing large datasets simultaneously by replicating identical computations on many processors. One can create enormously powerful parallel processors (at about 1/100th the cost of supercomputers), called multicomputers, which possess the one essential element supercomputers do not: scalability. Multicomputers do not have an asymptotic limit in performance and speed. They can be easily organized to provide greater processing capability; you simply connect more processors together. The scalable nature of multicomputers is a pervasive notion that escaped the authors of The Federal High Performance Computing Initiative. This oversight exposes the protectionist fundamentalism at work. Preserving a diminishing supercomputer industry by failing to acknowledge or support the emergence of a newer, more powerful technology is an act of covert institutionalized violence. Unfortunately, multicomputers alone will not satisfy the pressing need to build more complex simulations. Multicomputers depend on scalable software to work correctly. Little of this software has been written. Almost all of the software written in this country is not scalable. This "dusty deck" software must be completely rewritten to be effective on multicomputers. Rewriting this software is a costly activity that few organizations, let alone the Federal government, are willing to underwrite. The Initiative addresses software for supercomputer development only, not multicomputers. Supercomputers can be sold so long as "dusty deck" software is written for them. No multicomputers will be sold or produced until scalable software is written. Why not invest in scalable software simulations? The Dreams of Reason by Heinz Pagels posits the need to explore highly complex and nonlinear phenomena by reliance on multicomputer technology. Pagels was the first to recognize that complex systems are scalable. The economy, for instance, is a complex, scalable system. So is the human immunodeficiency virus (HIV), the Library of Congress, and the National Aerospace Plane. Pagels believes that simulating complex phenomena will improve the technological and economic growth of the United States, or without it, lead to a stagnation of intellect, problem solving, and vitality. This nation leads the world in software development activities. It is the one area of technology where the Japanese have made little progress. Software engineering is a highly creative process where personal initiative outstrips mechanical production line techniques. For multicomputers to really take off, engineers and scientists must be educated with the knowledge to build scalable software. Sadly, few universities in this country teach curriculum geared toward parallel computation. Enrollment in computer science curriculums are dropping. Multicomputer software and hardware education is prominent in both Japan and Europe. To advance the state of computing literacy and competence, the Bush Administration should launch a program targeting the construction of scalable software to support the next generation of computing and industrial infrastructure. This undertaking should include software for chemical reactions, structural dynamics, fluid dynamics, molecular and protein dynamics, multimedia systems, computer algebra, and scientific visualization. These subjects form the basis for technological innovation. If carefully planned and funded, this Strategic Software Initiative will give the U.S. a running start into the next century. Scalable multicomputer systems and computer simulations will give this nation a clear lead in societal productivity and intellectual prowess. Without them, we are doomed to second rate status. Caption: Richard M. Stein is a software engineer and writer from Pacific Palisades CA. His first book "Real-time Multicomputer Software Systems" will be published by Ellis Horwood Ltd. -- Richard M. Stein (aka, Rick 'Transputer' Stein) Sole proprietor of Rick's Software Toxic Waste Dump and Kitty Litter Co. "You build 'em, we bury 'em." uucp: ...{spsd, zardoz, felix}!dhw68k!stein