Information processing in a supercomputer

Abstract

The speed of today's computers range from the home computers, which can perform several hundred operations per second, to the powerful machines known as supercomputers, which have the capability of reaching peak computing speeds of over 100 million operations per second [12]. The first commercial electronic computer, UNIVAC I, was delivered to the Census Bureau in 1951. That computer was about three times faster than today's home computer and thousands of times more massive. Since the 1950's the speed of large scale scientific computers has doubled on the average every two years [12]. A major share of the increase in speed is attributable to inherently faster electronic parts. The earliest electronic computers used vacuum tubes for logic functions and magnetic drums for primary memory. Significant speed increases came when drums were replaced by magnetic cores, when vacuum tubes were replaced by transistors, and when both transistors and cores were replaced by integrated components [17, Ch. 8]. New technologies continually improved switching speeds as well as decreased propagation delay through miniaturization techniques.However, as improvement in switching devices and miniaturization reaches a limit imposed by optical resolution factors, it is apparent that any further significant increase in processing speed can be obtained only by the concurrent processing of a number of words [6]. Early computers approached a computational problem as a sequence of elementary steps to be executed one at a time, whereas supercomputers can process multiple elementary steps simultaneously. Thus computer architectural innovations which enable a machine to perform operations in parallel become extremely significant. (The term architecture refers to the logical organization of the computer as it is seen by the programmer [12].)Various supercomputer architectures have been proposed and constructed. Among these, three specific organizations are the parallel, pipeline and multiprocessing systems. After presenting some general information about supercomputers, this paper will survey these architectural types and give detailed examples of how specific supercomputers process information.