A REAL-TIME, INTERACTIVE, MU LTI PLE COMPUTER SYS EM R. B. McDowell P. F. Bohn N. K. Brown P. M. Kirk A. G. Witte Introduction HE COMPUTER, IN THE PAST DECADE, has grown function generation of the aerodynamics along from a device normally employed to manipu­ with the computation of trajectory, while the ana­ lateT numbers to a virtual automaton capable of log computer was to handle the rather high data­ mechanizing a very wide variety of processes. As rate, inherently analog functions involved in the the variety of applications grew, it became neces­ control system. The high precision and easy func­ sary for some applications to tie several computers tion generation of the digital computer could then, together. The connecting together of computers in combination with the high speed of the analog and the use of increasingly versatile terminals with machine, provide a single real-time simulation of these computers seems now to be replacing the all major aspects of a missile system. main trend of the 1950's and early 1960's, which The key problem in combining the two com­ was toward larger computers. Large computers puters was the design of linkage equipment to are still essential, and indeed are often the key to match the input-output characteristics of the two the interconnection of smaller computers and ter­ computers, including the control signals necessary minals; but the ability to interconnect is now a for proper timing of each machine's operations. dominant characteristic in the present general ex­ The first linkages or interfaces were designed pansion of the versatility of computers. in 1955 by EPSCO, Inc. for Ramo-Wooldridge Probably the earliest concept of interconnected Corp. (later Space Technology Laboratories) and computers consisted of an analog computer tied to Convair Astronautics. Key problems soon became a digital computer to form a hybridl computer. apparent: The original rationale for such a system was to 1. Too little control was centralized in the provide a real-time simulation * of a guided mis­ easily programmable digital computer and too sile, including aerodynamics, a long high-precision many unsupervised all-hardware analog opera­ trajectory, and the analog control system. The tions had to stay in precise voltage and timing digital computer was to handle the complicated adjustment for long periods of time. 2. The digital computer was too slow to keep 1 For a comprehensive description of hybrid computers and their applications see the book H ybrid Computation, by Bekey and up with the analog computer in many problems, Karplus, John Wiley & Sons, New York, 1968. and in any case had to be dedicated to handling * Simulation is the analytical or empirical modeling (analogy) of a theoretical or actual physical system. one program at a time. 2 APL Technical D igest Using existing technology, generalizing and extending it, a large real-time, interactive, multiple computer system has been developed. It is believed that this is the first time that real-time, interactive, and multiprogramming features have been successfully combined in a genera/­ purpose computer network. The work done has been oriented to a major extension of the use of computers for simulation. However, much of the interface and software design and implementation are sufficiently generalized to have many applications in systems other than simulation. 3. The use of vacuum tubes, along with too years with the key role of hybrid computers to little error checking in the hardware and software, simulate the Apollo guidance, control, aerody­ resulted in poor reliability and difficult mainte­ namics, etc. These massive simulations were used nance. The linkage, for example, required nine primarily for analysis, design, and testing before racks and contained several thousand vacuum and during hardware development. tubes. A second, and initially somewhat less conspic­ 4. The analog computers required a device uous aspect of computer interconnection also be­ controller and redesigned devices such as auto­ gan to be developed during the 1950's. This was set potentiometers, programmable logic, function the use of terminal devices located remotely, 2 switch control, automatic digital voltmeter addres­ wherever the user might be located, but tied to a sing, etc. in addition to the mode control and data central digital computer. The impetus for such transfer operations that were initially incorpo­ terminals grew from three needs: rated. 1. To extend standard computer peripheral 5. A way was required for the operator/ pro­ functions beyond the central computation center, grammer to communicate on-line with both com­ so that printing and program submission would be puters to modify his program or obtain variou more conveniently available. related computer services. 6. Efficient use of the digital computer re­ 2. To provide on-line conversational features quired a way for non-real-time jobs to run during to the user, minimizing turnaround time and the the increments of central processing unit time not user's dependence on computing center personnel. required by the real-time job. Such terminals as typewriters and graphics devices characterize these terminals. Despite these formidable limitations, both the Ramo-Wooldridge and Convair systems were used 3. To allow direct data acquisition and/ or pro­ (albeit laboriously) to successfully solve several cess control at remote sites to eliminate the need aerospace problems. As a consequence, an appre­ for separate, dedicated computers or the awkward ciation of what might be accomplished by a hy­ brid computer system spread rapidly within the 2 An excellent text on this technology in terms of the hardware is the book Telecommunications and the Computer; by James aerospace industry, culminating in the last severa] Martin, Prentice-Hall, Englewood Cliffs, N.J., 1969. November - December 1971 3 and time-consuming manual transfer of data by burden be placed on the interface for each device way of magnetic or paper tapc or by punched to be interconnected in order to minimize and cards. hopefully eliminate hardware or software design The terminals used to meet these needs are of changes to previously connected devices. a wide variety, but fall into two general classifica­ 5. The devclopment of terminal support soft­ tions: (a) noncomputer or "nonintelligent" de­ ware sufficient to allow utilization of all hardware vices and (b) computer or "intelligent" devices. features of each interconnected device. On-line The basic difference between the two types is, of operator / programmer communication with exe­ course, whether or not they have a general (usu­ cuting programs is a key feature of much of this ally programmable) computation capability. Mini­ type of software. computers, t which have recently come into prom­ 6. The development, as a part of the basic inence, are finding a major application as the key supervisory software operating system, of such clement in the so-called intelligent terminal. Type­ features as multiprogramming, real-time services, writers are far and away the largest category and multiprocessingt which automatically adjust within the nonintelligent terminal classification. the computer operations to the changing demands Finally, a last and rather recent class of com­ on the overall system. These features are aimed at puter interconnection is that of the digital com­ maximizing the efficiency of the total system, while puter network where several computers, often at the same time providing each user with what quite large ones, are connected together to pro­ appears to be a computer dedicated to his partic­ vide immediate access to a comprehensive data ular job. base. Usually such systems also provide extensive, A seventh key technical problem area for inter­ though as yet rather specialized, computational connected systems in general is reliable, long-dis­ and other data handling services. tance, high-speed data communications. This, All these types of system, however, must find however, was not a major factor for the systems solutions to the same basic hardware and software discussed in this paper. problems associated with the development of the hybrid computer. At this point, it will be useful Development Context to reorganize and restate these technical problems During 1967 and 1968 the Laboratory identi­ as follows: fied four primary areas where simulation require­ 1. The development of comprehensive software ments exceeded existing capabilities: supervisory services to allocate or manage, as well 1. Hybrid computing. as error-check, all hardware and software opera­ 2. Interactive, all-digital simulation coupled tions. IBM terms this the operating system. Others with operating hardware. variously call it the monitor, supervisor, etc. 3. Missile and radar receiver simulation. 2. The development of parallel hardware sub­ 4. Radar system evaluation using realistic com­ systems, particularly for input-output (I/ O) chan­ puter-generated target scenarios. nels to speed up key functions by making them Such needs usually had been handled by sepa­ less dependent on the computer's central process­ rate, stand-alone, often special-purpose computers. ing unit (CPU). In order to gain projected economies of scale in 3. The use of solid-state components in place terms of cost per unit work, and to provide a gen­ of vacuum tubes and electromechanical devices to eralized system capability more adaptable to fu­ increase both speed and reliability. ture needs, a decision was made to develop a 4. The development