Biograpny Citation for the Reginald Fessenden Award
Contributed by Carl Savit
Over the years devices or ideas have been introduced that have achieved wide-spread currency. In many cases the origins of those devices and ideas are obscured. The general feeling is that all sorts of people have had a hand in their creation, that the devices or ideas "have always been around." Just such a device is the computer peripheral known as the array processor. Fortunately, however, it is possible to attribute the origin of the first true array processor, the IBM 2938, to John Koonce and his partner Byron Gariepy who is also being honored this evening.
John and Byron originated and developed the concept of a multi-instruction, floating-point, vector processor to be attached to a central processing computer and approached us at Western Geophysical Company to specify its instruction set and other characteristics. By 1967, the prototype device had been delivered and installed. From 1967 until about 1974, no competing devices appeared on the market despite the enormous advance in geophysical data processing power that such a peripheral afforded. After that, array processors became available to fit every need and every budget.
Today, virtually all geophysical processing computers either have one or more peripheral array processors, or have a built-in vector processor as in the so-called supercomputers. In today's seismic data processing, the central processing unit or the scalar part of a supercomputer is confined to traffic control, sorting, and the calculation of processing parameters. All the actual operations on data are performed in the array processors or the vector part of the unit.
John Koonce's penchant for innovation may have been inborn, but he had an excellent education at Cornell and Southern Methodist Universities, receiving a B.S. with high honors in mechanical engineering from the latter institution in 1962. Employed by IBM as a system engineer in Shreveport, Louisiana, he was assigned to work with Western's first computer center on its establishment in that city in 1964. He wasted no time in determining that the major bottleneck to efficient processing of seismic data was the requirement for identical, repetitive operations, usually involving multiplications, to long columns of individual data values. The most primitive attack on the problem had been to produce a fast multiply-add instruction. But what clearly was needed was a device that could execute many if not all the operations that must be applied to individual data values. With a set of 12 instructions, John and Byron's device came near to that goal. Among the optional instructions that were offered was the first hardware implemented fast Fourier transform, the ubiquitous workhorse of all subsequent array processors and indispensible in modern data processing.
Vastly more important than the immediate results of their development was the inspiration to emulate and improve on their pioneering efforts. If not for the development of a whole host of array processors and vector machines, seismic data processing as we know it today would not be possible. Processing costs would have to be at least 100 times as great as they are now.
The Reginald Fessenden Award is given "to a person who has made a specific technical contribution to exploration geophysics." It is difficult to imagine a greater such contribution than that of John Koonce.