Architecture (Несколько текстов для зачёта), страница 2

Computer scientists continue to develop new algorithms and data structures to solve new problems and improve the efficiency of existing programs. One area of theoretical research is called algorithmic complexity. Computer scientists in this field seek to develop techniques for determining the inherent efficiency of algorithms with respect to one another. Another area of theoretical research called computability theory seeks to identify the inherent limits of computation.

Software engineers use programming languages to communicate algorithms to a computer. Natural languages such as English are ambiguous—meaning that their grammatical structure and vocabulary can be interpreted in multiple ways—so they are not suited for programming. Instead, simple and unambiguous artificial languages are used. Computer scientists study ways of making programming languages more expressive, thereby simplifying programming and reducing errors. A program written in a programming language must be translated into machine language (the actual instructions that the computer follows). Computer scientists also develop better translation algorithms that produce more efficient machine language programs.

Databases and information retrieval are related fields of research. A database is an organized collection of information stored in a computer, such as a company’s customer account data. Computer scientists attempt to make it easier for users to access databases, prevent access by unauthorized users, and improve access speed. They are also interested in developing techniques to compress the data, so that more can be stored in the same amount of memory. Databases are sometimes distributed over multiple computers that update the data simultaneously, which can lead to inconsistency in the stored information. To address this problem, computer scientists also study ways of preventing inconsistency without reducing access speed.

Information retrieval is concerned with locating data in collections that are not clearly organized, such as a file of newspaper articles. Computer scientists develop algorithms for creating indexes of the data. Once the information is indexed, techniques developed for databases can be used to organize it. Data mining is a closely related field in which a large body of information is analyzed to identify patterns. For example, mining the sales records from a grocery store could identify shopping patterns to help guide the store in stocking its shelves more effectively.

Operating systems are programs that control the overall functioning of a computer. They provide the user interface, place programs into the computer’s memory and cause it to execute them, control the computer’s input and output devices, manage the computer’s resources such as its disk space, protect the computer from unauthorized use, and keep stored data secure. Computer scientists are interested in making operating systems easier to use, more secure, and more efficient by developing new user interface designs, designing new mechanisms that allow data to be shared while preventing access to sensitive data, and developing algorithms that make more effective use of the computer’s time and memory.

The study of numerical computation involves the development of algorithms for calculations, often on large sets of data or with high precision. Because many of these computations may take days or months to execute, computer scientists are interested in making the calculations as efficient as possible. They also explore ways to increase the numerical precision of computations, which can have such effects as improving the accuracy of a weather forecast. The goals of improving efficiency and precision often conflict, with greater efficiency being obtained at the cost of precision and vice versa.

Symbolic computation involves programs that manipulate nonnumeric symbols, such as characters, words, drawings, algebraic expressions, encrypted data (data coded to prevent unauthorized access), and the parts of data structures that represent relationships between values. One unifying property of symbolic programs is that they often lack the regular patterns of processing found in many numerical computations. Such irregularities present computer scientists with special challenges in creating theoretical models of a program’s efficiency, in translating it into an efficient machine language program, and in specifying and testing its correct behavior.

Computer architecture is the design and analysis of new computer systems. Computer architects study ways of improving computers by increasing their speed, storage capacity, and reliability, and by reducing their cost and power consumption. Computer architects develop both software and hardware models to analyze the performance of existing and proposed computer designs, then use this analysis to guide development of new computers. They are often involved with the engineering of a new computer because the accuracy of their models depends on the design of the computer’s circuitry. Many computer architects are interested in developing computers that are specialized for particular applications such as image processing, signal processing, or the control of mechanical systems. The optimization of computer architecture to specific tasks often yields higher performance, lower cost, or both.

Artificial intelligence (AI) research seeks to enable computers and machines to mimic human intelligence and sensory processing ability, and models human behavior with computers to improve our understanding of intelligence. The many branches of AI research include machine learning, inference, cognition, knowledge representation, problem solving, case-based reasoning, natural language understanding, speech recognition, computer vision, and artificial neural networks.

A key technique developed in the study of artificial intelligence is to specify a problem as a set of states, some of which are solutions, and then search for solution states. For example, in chess, each move creates a new state. If a computer searched the states resulting from all possible sequences of moves, it could identify those that win the game. However, the number of states associated with many problems (such as the possible number of moves needed to win a chess game) is so vast that exhaustively searching them is impractical. The search process can be improved through the use of heuristics—rules that are specific to a given problem and can therefore help guide the search. For example, a chess heuristic might indicate that when a move results in checkmate, there is no point in examining alternate moves.

Another area of computer science that has found wide practical use is robotics—the design and development of computer controlled mechanical devices. Robots range in complexity from toys to automated factory assembly lines, and relieve humans from tedious, repetitive, or dangerous tasks. Robots are also employed where requirements of speed, precision, consistency, or cleanliness exceed what humans can accomplish. Roboticists—scientists involved in the field of robotics—study the many aspects of controlling robots. These aspects include modeling the robot’s physical properties, modeling its environment, planning its actions, directing its mechanisms efficiently, using sensors to provide feedback to the controlling program, and ensuring the safety of its behavior. They also study ways of simplifying the creation of control programs. One area of research seeks to provide robots with more of the dexterity and adaptability of humans, and is closely associated with AI.

Human-computer interfaces provide the means for people to use computers. An example of a human-computer interface is the keyboard, which lets humans enter commands into a computer and enter text into a specific application. The diversity of research into human-computer interfacing corresponds to the diversity of computer users and applications. However, a unifying theme is the development of better interfaces and experimental evaluation of their effectiveness. Examples include improving computer access for people with disabilities, simplifying program use, developing three-dimensional input and output devices for virtual reality, improving handwriting and speech recognition, and developing heads-up displays for aircraft instruments in which critical information such as speed, altitude, and heading are displayed on a screen in front of the pilot’s window. One area of research, called visualization, is concerned with graphically presenting large amounts of data so that people can comprehend its key properties.

Because computer science grew out of mathematics and electrical engineering, it retains many close connections to those disciplines. Theoretical computer science draws many of its approaches from mathematics and logic. Research in numerical computation overlaps with mathematics research in numerical analysis. Computer architects work closely with the electrical engineers who design the circuits of a computer.

Beyond these historical connections, there are strong ties between AI research and psychology, neurophysiology, and linguistics. Human-computer interface research also has connections with psychology. Roboticists work with both mechanical engineers and physiologists in designing new robots.

Computer science also has indirect relationships with virtually all disciplines that use computers. Applications developed in other fields often involve collaboration with computer scientists, who contribute their knowledge of algorithms, data structures, software engineering, and existing technology. In return, the computer scientists have the opportunity to observe novel applications of computers, from which they gain a deeper insight into their use. These relationships make computer science a highly interdisciplinary field of study.

Parallel Processing

Parallel Processing, computer technique in which multiple operations are carried out simultaneously. Parallelism reduces computational time. For this reason, it is used for many computationally intensive applications such as predicting economic trends or generating visual special effects for feature films.

Two common ways that parallel processing is accomplished are through multiprocessing or instruction-level parallelism. Multiprocessing links several processors—computers or microprocessors (the electronic circuits that provide the computational power and control of computers)—together to solve a single problem. Instruction-level parallelism uses a single computer processor that executes multiple instructions simultaneously.

If a problem is divided evenly into ten independent parts that are solved simultaneously on ten computers, then the solution requires one tenth of the time it would take on a single nonparallel computer where each part is solved in sequential order. Many large problems are easily divisible for parallel processing; however, some problems are difficult to divide because their parts are interdependent, requiring the results from another part of the problem before they can be solved.

Portions of a problem that cannot be calculated in parallel are called serial. These serial portions determine the computation time for a problem. For example, suppose a problem has nine million computations that can be done in parallel and one million computations that must be done serially. Theoretically, nine million computers could perform nine-tenths of the total computation simultaneously, leaving one-tenth of the total problem to be computed serially. Therefore, the total execution time is only one-tenth of what it would be on a single nonparallel computer, despite the additional nine million processors.

In 1966 American electrical engineer Michael Flynn distinguished four classes of processor architecture (the design of how processors manipulate data and instructions). Data can be sent either to a computer's processor one at a time, in a single data stream, or several pieces of data can be sent at the same time, in multiple data streams. Similarly, instructions can be carried out either one at a time, in a single instruction stream, or several instructions can be carried out simultaneously, in multiple instruction streams.

Serial computers have a Single Instruction stream, Single Data stream (SISD) architecture. One piece of data is sent to one processor. For example, if 100 numbers had to be multiplied by the number 3, each number would be sent to the processor, multiplied, and the result stored; then the next number would be sent and calculated, until all 100 results were calculated. Applications that are suited for SISD architectures include those that require complex interdependent decisions, such as word processing.

A Multiple Instruction stream, Single Data stream (MISD) processor replicates a stream of data and sends it to multiple processors, each of which then executes a separate program. For example, the contents of a database could be sent simultaneously to several processors, each of which would search for a different value. Problems well-suited to MISD parallel processing include computer vision systems that extract multiple features, such as vegetation, geological features, or manufactured objects, from a single satellite image.