# Supercomputers in the office.

Supercomputers In the Office

Solving business problems with science

Supercomputers are used to solve computationally difficult problems in science and engineering: the study of wind flows over an aircraft wing, analysis of the reaction kinetics of a new pharmaceutical compound, and deciding where to drill for oil based upon seismic data analysis. The power of these machines to perform hundreds of millions of arithmetic steps per second makes them ideal tools for uncovering and understanding the complex nature of our physical world.

With increasing frequency, supercomputers are starting to be used in the commercial world to increase the efficiency and productivity of business operations and planning. High-performance computing is used in this environment not to reveal the secrets of nature, but to better analyze and understand complex systems made by man. Moreover, while complex natural phenomena differ in many respects from complex man-made systems, both can be analyzed and better understood using supercomputer technology and sophisticated computational techniques.

COMPUTATIONAL EXPERIMENTATION

Historically, science has been divided into two categories: theory and experimentation. Theoreticians first derive their views of the world from principles and are secondarily concerned with whether or not their theories are confirmable in the real world. Experimentalists, on the other hand, have faith in phenomena that can be demonstrated in the laboratory, regardless of the adherence to theory. This categorization is, of course, extreme, but it represents the poles of a spectrum.

The supercomputer has given rise to a third branch of science: computational experimentation. Some experiments are too expensive or time consuming to be attempted (for example, studying wind effects over an entire jumbo jet), and some are impossible (large-scale nuclear weapons testing and studying fictitious chemical compounds). In these cases, computationalists build a mathematical model of the system to be studied and then use a supercomputer to carry out the experiment. This technique allows faster and less expensive inquiry and vicarious trial and error of different system scenarios.

A mathematical model captures the physical attributes of a complex system as an interwoven set of mathematical relationships. The model can then be operated by a supercomputer to analyze or simulate the operation of the system. These tools from the physical sciences--mathematical models and supercomputers--are increasingly being used in commercial environments to study the functioning of complex business systems.

COMPLEX BUSINESS SYSTEMS

Many business enterprises encounter complex problems that are candidates for computational methods and supercomputing. Examples are prioritizing and selecting alternative courses of action for logistical and scheduling problems in communications and manufacturing, such as the minimum-cost selection of vendors for subcomponent manufacturing; the optimal placement of concentrators in a communication network; scheduling the most cost-effective distribution of finished goods to meet customer demand; and real-time rerouting around failures in a telephone or power generation network.

The proliferation of desktop personal computers in the office has generated interest in this type problem. For the first time, technicians and managers have ready access to sophisticated computational tools. Indeed, moderate priced PC-class machines are capable of solving small versions of operational and planning problems that have significant numeric content. For more realistic, full-sized, complex problems, however, personal computers often lack the computational power to provide solutions within a reasonable time. As these problems become larger, the corresponding computational power requirements often increase at a geometric rate. Problems that accurately capture and model the details of a particular business function often require the computational sophistication of supercomputing.

The transportation industry contains a variety of complex business systems that can be studied effectively using supercomputer technology. Examples include railroad train scheduling and routing, truck dispatching and routing, and the interconnections between railroads, trucking, and shipping. Deregulation in many segments of the transportation industry has significantly increased the need to better understand how these systems operate and to increase utilization and productivity. Slight improvements in operational efficiency can make the difference in an environment with fierce competition and thin profit margins.

Nowhere is this complex business environment more evident than in the airline industry. During the last decade, airlines have been buffeted by deregulation and consolidation. The profitable companies that have survived have become more efficient and streamlined, striving to get the most from operational expenditures. Not surprisingly, airline companies are leading the vanguard in the use of supercomputer technology to increase productivity. Computer-intensive applications include fleet scheduling, maintenance tracking and scheduling, crew and personnel assignment, and system-wide recovery from weather disruptions. The analysis and understanding of the complexity and interdependency of these systems demand the power of a supercomputer and the ingenuity of sophisticated mathematical techniques.

Another sector that has seen major restructuring in the past decade is the financial industry. The securities industry, for example, has had to adjust to worldwide around-the-clock trading, significant government deregulation, and destabilizing international currency fluctuations. The net effect has been to induce new levels of chaos into an already chaotic environment. In many cases, the response has been to attempt better understanding of basic underlying economic phenomena using new mathematical and forecasting techniques. Because conditions change fast in this environment--people often want answers to complex problems in seconds or minutes--the computational power offered by supercomputers is being recognized and exploited. Financial applications that can use supercomputer technology include analysis of currency exchange cycles, optimal portfolio selection to minimize risk, and the analysis of financial futures versus underlying financial instruments.

SUPERCOMPUTER TECHNOLOGY AND METHODOLOGY

The definition of "supercomputer" is an elusive concept. Many computer companies are labeling products "mini-super," "super-mini," "disk-top-super," and other hyphenated nicknames. As this article was being written, the U.S. Department of Commerce was wrestling with the quantitative definition of supercomputer for purposes of export control. As technology speeds ahead, the definition of supercomputer will continue changing.

Qualitatively, supercomputers are those machines at the leading edge of technology in terms of computing power and the overall ability to perform many computing jobs very fast. For practical purposes, this definition is adequate for the discussion of supercomputers in business.

The computational methods most often used with supercomputers for the analysis of complex business problems are mathematical optimization and mathematical simulation. Mathematical optimization is a technique for systematically deciding how to best choose or expend scarce resources in the presence of competing demands. Optimization problems often arise for decision making in the face of large numbers of alternatives; an optimization technique analyzes the different available courses of action and chooses the best one based upon some business criterion (e.g., minimal cost, maximum profit, least disruption to existing operations, etc.).

A classic optimization problem is the airline crew scheduling problem: all flights made by an airline must be assigned cockpit and cabin personnel so as to meet logistical and contractual conditions and to minimize total labor costs. Even for a moderate-sized airline fleet, the alternatives for assigning crews to flights number in the millions; effectively dealing with this problem requires many hours of computer time, even for state-of-the-art supercomputers.

Using optimization involves building a mathematical model of the real-world process or system under consideration and analyzing the model using mathematical techniques to determine the best course of action for the real-world process or system. The modeling process--translating real-world relationships among elements of a business system into mathematical relationships in the model--is an intellectual problem requiring people with a fundamental knowledge of the business. After representative mathematical models have been constructed by people, computers are very good at analyzing them, and supercomputers analyze them very fast.

Mathematical simulation also involves building a model of a real-world system using mathematical relationships to capture the workings of the system. In simulation, the model mimics the operation of the system overtime. Studying how the model reacts to varying internal and external influences over a particular simulated time interval gives insight into how the real system will operate under varying conditions. For example, the interaction of foreign exchange rates can be modeled using Monte Carlo simulation. Exchange rates for various currencies are set at arbitrary random values in the model and then allowed to fluctuate and interact over time. Repeating this scenario thousands of times with random starting values and then statistically analyzing the results can give insight into the real system of international currency rate interaction and fluctuation.

SUPERCOMPUTER BENEFITS: INSIGHT AND KNOWLEDGE

Computational experimentation using supercomputers offers a new tool for analyzing and understanding the operation of complex business processes. Applying supercomputer technology significantly compresses the amount of time required to perform complex calculations, allowing existing compute-intensive applications to be operated much more quickly. Operational and strategic scheduling scenarios that require many minutes or hours to compute on a departmental minicomputer can be computed in seconds or minutes on a supercomputer. This compression of compute time for analyzing a process leads to new ways of thinking about problems.

More importantly, the significant increase in the amount of computer power offered by supercomputers allows people to attack problems and implement applications that previously would not have been attempted. People naturally shy away from even formulating a problem if they know the problem is intractable. Problems that were previously considered unsolvable --the airline crew selection problem mentioned previously and the job-shop scheduling problem from manufacturing--are being successfully attacked using supercomputer technology and sophisticated computational techniques.

The diagram on page 5 demonstrates how people's thinking changes when significant computing power is applied to a specific problem. The diagram illustrates the quantitative and qualitative effects of a 200-fold speed-up in the time required to obtain the solution to a problem--about the order of increase in computing power going from a departmental minicomputer to a supercomputer. A problem that would previously have required months or years to compute--clearly beyond the capacity and patience of all but the most tenacious investigator--is reduced to tens of hours. This is an example of an intractable problem entering the realm of feasibility, possibly yielding new knowledge about some process or system under study.

A problem that previously required hours or days to compute is reduced to minutes, allowing more cases to be studied; more scenarios to be run; and, in general, more insight and knowledge to be gleaned about the underlying problem. And finally, applications that previously required seconds or minutes to compute are performed almost instantaneously, allowing more people in the enterprise to interact with the computer application, resulting in wider access to the resulting knowledge.

THE CONVERGENCE OF SCIENTIFIC AND COMMERCIAL COMPUTING

The use of computers by scientists and business people has historically been quite different. Scientists use computers to probe the secrets of the physical world, translating mathematical formulas into computer languages, consuming large amounts of supercomputer time for repetitive mathematical calculations, and displaying results as charts, graphs and images on computer workstations. Business people view computers as expensive but effective bookkeepers, capable of tracking and maintaining accounts receivable, performing periodic payroll functions, and perhaps a few chores of a computational analytic nature such as inventory control. The computer tools of interest to the business data processing manager are disk drives for storing the enterprise database and high-speed printers for producing reports.

With the infiltration of supercomputers into the workplace, these different approaches to computers and computing are converging. Scientists are becoming aware that the computational study of chaos and instability in physical systems is remarkably similar to the study of these phenomena in complex business systems, and, in many cases, the same tools and techniques apply in both domains.

Likewise, business people see that some aspects of modern business operations are too complex and extensive to understand and manage using traditional methods, and that scientific analysis and simulation techniques provide valuable insight and knowledge that lead to increased management effectiveness and overall enterprise productivity.

It is illustrative that many of the computational analysts in Wall Street security firms--affectionately known to their colleagues as "Rocket Scientists"--have, in many cases, their basic education, training, and experience in the physical sciences. The movement of stock prices on the New York Stock Exchange at various times during the day and the movement of molecules in a solution at various temperatures appear to be totally unrelated problems. At a fundamental level, however, both can be studied and insight provided using similar computational tools, techniques, and technologies.

PHOTO : Financial applications that can use supercomputer technology include analysis of currency

PHOTO : exchange cycles, optimal portfolio selection to minimize risk, and the analysis of

PHOTO : financial futures versus underlying financial instruments.

PHOTO : The airline industry uses supercomputers for fleet scheduling, for maintenance tracking

PHOTO : and scheduling, and for crew and personnel assignments. Frito-Lay incorporates

PHOTO : supercomputers for optimizing plant floor operations and production.

Mr. Crowder is an application and technology consultant at the IBM Corporation in Palo Alto, California.

Solving business problems with science

Supercomputers are used to solve computationally difficult problems in science and engineering: the study of wind flows over an aircraft wing, analysis of the reaction kinetics of a new pharmaceutical compound, and deciding where to drill for oil based upon seismic data analysis. The power of these machines to perform hundreds of millions of arithmetic steps per second makes them ideal tools for uncovering and understanding the complex nature of our physical world.

With increasing frequency, supercomputers are starting to be used in the commercial world to increase the efficiency and productivity of business operations and planning. High-performance computing is used in this environment not to reveal the secrets of nature, but to better analyze and understand complex systems made by man. Moreover, while complex natural phenomena differ in many respects from complex man-made systems, both can be analyzed and better understood using supercomputer technology and sophisticated computational techniques.

COMPUTATIONAL EXPERIMENTATION

Historically, science has been divided into two categories: theory and experimentation. Theoreticians first derive their views of the world from principles and are secondarily concerned with whether or not their theories are confirmable in the real world. Experimentalists, on the other hand, have faith in phenomena that can be demonstrated in the laboratory, regardless of the adherence to theory. This categorization is, of course, extreme, but it represents the poles of a spectrum.

The supercomputer has given rise to a third branch of science: computational experimentation. Some experiments are too expensive or time consuming to be attempted (for example, studying wind effects over an entire jumbo jet), and some are impossible (large-scale nuclear weapons testing and studying fictitious chemical compounds). In these cases, computationalists build a mathematical model of the system to be studied and then use a supercomputer to carry out the experiment. This technique allows faster and less expensive inquiry and vicarious trial and error of different system scenarios.

A mathematical model captures the physical attributes of a complex system as an interwoven set of mathematical relationships. The model can then be operated by a supercomputer to analyze or simulate the operation of the system. These tools from the physical sciences--mathematical models and supercomputers--are increasingly being used in commercial environments to study the functioning of complex business systems.

COMPLEX BUSINESS SYSTEMS

Many business enterprises encounter complex problems that are candidates for computational methods and supercomputing. Examples are prioritizing and selecting alternative courses of action for logistical and scheduling problems in communications and manufacturing, such as the minimum-cost selection of vendors for subcomponent manufacturing; the optimal placement of concentrators in a communication network; scheduling the most cost-effective distribution of finished goods to meet customer demand; and real-time rerouting around failures in a telephone or power generation network.

The proliferation of desktop personal computers in the office has generated interest in this type problem. For the first time, technicians and managers have ready access to sophisticated computational tools. Indeed, moderate priced PC-class machines are capable of solving small versions of operational and planning problems that have significant numeric content. For more realistic, full-sized, complex problems, however, personal computers often lack the computational power to provide solutions within a reasonable time. As these problems become larger, the corresponding computational power requirements often increase at a geometric rate. Problems that accurately capture and model the details of a particular business function often require the computational sophistication of supercomputing.

The transportation industry contains a variety of complex business systems that can be studied effectively using supercomputer technology. Examples include railroad train scheduling and routing, truck dispatching and routing, and the interconnections between railroads, trucking, and shipping. Deregulation in many segments of the transportation industry has significantly increased the need to better understand how these systems operate and to increase utilization and productivity. Slight improvements in operational efficiency can make the difference in an environment with fierce competition and thin profit margins.

Nowhere is this complex business environment more evident than in the airline industry. During the last decade, airlines have been buffeted by deregulation and consolidation. The profitable companies that have survived have become more efficient and streamlined, striving to get the most from operational expenditures. Not surprisingly, airline companies are leading the vanguard in the use of supercomputer technology to increase productivity. Computer-intensive applications include fleet scheduling, maintenance tracking and scheduling, crew and personnel assignment, and system-wide recovery from weather disruptions. The analysis and understanding of the complexity and interdependency of these systems demand the power of a supercomputer and the ingenuity of sophisticated mathematical techniques.

Another sector that has seen major restructuring in the past decade is the financial industry. The securities industry, for example, has had to adjust to worldwide around-the-clock trading, significant government deregulation, and destabilizing international currency fluctuations. The net effect has been to induce new levels of chaos into an already chaotic environment. In many cases, the response has been to attempt better understanding of basic underlying economic phenomena using new mathematical and forecasting techniques. Because conditions change fast in this environment--people often want answers to complex problems in seconds or minutes--the computational power offered by supercomputers is being recognized and exploited. Financial applications that can use supercomputer technology include analysis of currency exchange cycles, optimal portfolio selection to minimize risk, and the analysis of financial futures versus underlying financial instruments.

SUPERCOMPUTER TECHNOLOGY AND METHODOLOGY

The definition of "supercomputer" is an elusive concept. Many computer companies are labeling products "mini-super," "super-mini," "disk-top-super," and other hyphenated nicknames. As this article was being written, the U.S. Department of Commerce was wrestling with the quantitative definition of supercomputer for purposes of export control. As technology speeds ahead, the definition of supercomputer will continue changing.

Qualitatively, supercomputers are those machines at the leading edge of technology in terms of computing power and the overall ability to perform many computing jobs very fast. For practical purposes, this definition is adequate for the discussion of supercomputers in business.

The computational methods most often used with supercomputers for the analysis of complex business problems are mathematical optimization and mathematical simulation. Mathematical optimization is a technique for systematically deciding how to best choose or expend scarce resources in the presence of competing demands. Optimization problems often arise for decision making in the face of large numbers of alternatives; an optimization technique analyzes the different available courses of action and chooses the best one based upon some business criterion (e.g., minimal cost, maximum profit, least disruption to existing operations, etc.).

A classic optimization problem is the airline crew scheduling problem: all flights made by an airline must be assigned cockpit and cabin personnel so as to meet logistical and contractual conditions and to minimize total labor costs. Even for a moderate-sized airline fleet, the alternatives for assigning crews to flights number in the millions; effectively dealing with this problem requires many hours of computer time, even for state-of-the-art supercomputers.

Using optimization involves building a mathematical model of the real-world process or system under consideration and analyzing the model using mathematical techniques to determine the best course of action for the real-world process or system. The modeling process--translating real-world relationships among elements of a business system into mathematical relationships in the model--is an intellectual problem requiring people with a fundamental knowledge of the business. After representative mathematical models have been constructed by people, computers are very good at analyzing them, and supercomputers analyze them very fast.

Mathematical simulation also involves building a model of a real-world system using mathematical relationships to capture the workings of the system. In simulation, the model mimics the operation of the system overtime. Studying how the model reacts to varying internal and external influences over a particular simulated time interval gives insight into how the real system will operate under varying conditions. For example, the interaction of foreign exchange rates can be modeled using Monte Carlo simulation. Exchange rates for various currencies are set at arbitrary random values in the model and then allowed to fluctuate and interact over time. Repeating this scenario thousands of times with random starting values and then statistically analyzing the results can give insight into the real system of international currency rate interaction and fluctuation.

SUPERCOMPUTER BENEFITS: INSIGHT AND KNOWLEDGE

Computational experimentation using supercomputers offers a new tool for analyzing and understanding the operation of complex business processes. Applying supercomputer technology significantly compresses the amount of time required to perform complex calculations, allowing existing compute-intensive applications to be operated much more quickly. Operational and strategic scheduling scenarios that require many minutes or hours to compute on a departmental minicomputer can be computed in seconds or minutes on a supercomputer. This compression of compute time for analyzing a process leads to new ways of thinking about problems.

More importantly, the significant increase in the amount of computer power offered by supercomputers allows people to attack problems and implement applications that previously would not have been attempted. People naturally shy away from even formulating a problem if they know the problem is intractable. Problems that were previously considered unsolvable --the airline crew selection problem mentioned previously and the job-shop scheduling problem from manufacturing--are being successfully attacked using supercomputer technology and sophisticated computational techniques.

The diagram on page 5 demonstrates how people's thinking changes when significant computing power is applied to a specific problem. The diagram illustrates the quantitative and qualitative effects of a 200-fold speed-up in the time required to obtain the solution to a problem--about the order of increase in computing power going from a departmental minicomputer to a supercomputer. A problem that would previously have required months or years to compute--clearly beyond the capacity and patience of all but the most tenacious investigator--is reduced to tens of hours. This is an example of an intractable problem entering the realm of feasibility, possibly yielding new knowledge about some process or system under study.

A problem that previously required hours or days to compute is reduced to minutes, allowing more cases to be studied; more scenarios to be run; and, in general, more insight and knowledge to be gleaned about the underlying problem. And finally, applications that previously required seconds or minutes to compute are performed almost instantaneously, allowing more people in the enterprise to interact with the computer application, resulting in wider access to the resulting knowledge.

THE CONVERGENCE OF SCIENTIFIC AND COMMERCIAL COMPUTING

The use of computers by scientists and business people has historically been quite different. Scientists use computers to probe the secrets of the physical world, translating mathematical formulas into computer languages, consuming large amounts of supercomputer time for repetitive mathematical calculations, and displaying results as charts, graphs and images on computer workstations. Business people view computers as expensive but effective bookkeepers, capable of tracking and maintaining accounts receivable, performing periodic payroll functions, and perhaps a few chores of a computational analytic nature such as inventory control. The computer tools of interest to the business data processing manager are disk drives for storing the enterprise database and high-speed printers for producing reports.

With the infiltration of supercomputers into the workplace, these different approaches to computers and computing are converging. Scientists are becoming aware that the computational study of chaos and instability in physical systems is remarkably similar to the study of these phenomena in complex business systems, and, in many cases, the same tools and techniques apply in both domains.

Likewise, business people see that some aspects of modern business operations are too complex and extensive to understand and manage using traditional methods, and that scientific analysis and simulation techniques provide valuable insight and knowledge that lead to increased management effectiveness and overall enterprise productivity.

It is illustrative that many of the computational analysts in Wall Street security firms--affectionately known to their colleagues as "Rocket Scientists"--have, in many cases, their basic education, training, and experience in the physical sciences. The movement of stock prices on the New York Stock Exchange at various times during the day and the movement of molecules in a solution at various temperatures appear to be totally unrelated problems. At a fundamental level, however, both can be studied and insight provided using similar computational tools, techniques, and technologies.

PHOTO : Financial applications that can use supercomputer technology include analysis of currency

PHOTO : exchange cycles, optimal portfolio selection to minimize risk, and the analysis of

PHOTO : financial futures versus underlying financial instruments.

PHOTO : The airline industry uses supercomputers for fleet scheduling, for maintenance tracking

PHOTO : and scheduling, and for crew and personnel assignments. Frito-Lay incorporates

PHOTO : supercomputers for optimizing plant floor operations and production.

Mr. Crowder is an application and technology consultant at the IBM Corporation in Palo Alto, California.

Printer friendly Cite/link Email Feedback | |

Author: | Crowder, Harlan P. |
---|---|

Publication: | Business Perspectives |

Date: | Mar 22, 1990 |

Words: | 2100 |

Previous Article: | Cheap energy and the economic boom. |

Next Article: | Computer integrated manufacturing. |

Topics: |