The simulation of time sharing systems

Abstract

The development of new large scale time-sharing systems has raised a number of problems for computation center management. Not only is it necessary to develop an appropriate hardware configuration for these systems, but appropriate software adjustments must be made. Unfortunately, these systems often do not respond to changes in the manner that intuition would suggest, and there are few guides to assist in the analysis of performance characteristics. The development of a comprehensive simulation model to assist in the investigation of these questions is described in this paper. The resulting model has a general purpose design and can be used to study a variety of time-sharing systems. It can also be used to assist in the design and development of new time-sharing algorithms or techniques. For the sake of efficiency and greater applicability, the model was implemented in a limited FORTRAN subset that is compatible with most FORTRAN IV compilers. The use of the simulation is demonstrated by a study of the IBM 360/67 tlme-sharing system. I n t r o d u c t i o nThe brief history of automatic digital computing has been marked by a series of hardware developments followed by striking changes in computer usage. At the present time we are midway between one of these new developments and the resulting change. The success of the experimental time-sharing systems which have been developed at such places as the Massachusetts Institute of Technology (Project MAC) [1, 2], the RAND Corporation [10,11], and the System Development Corporation [7][8][9] have clearly demonstrated both the desirability and the feasibility of this technique for computer utilization. Now, new third generation computers employing solid logic or integrated circuit (IC) technology are becoming available. Not only are these machines able to eliminate many of the hardware-imposed limitations of earlier timesharing systems, but certain models within some manufacturers' lines are being marketed especially for general purpose time-sharing applications. The first deliveries of some of the new hardware have already taken place, and comprehensive software packages are currently under development for these systems. Consequently, it is expected that a sharp change will take place over the next few years in the manner in which man employs the computer.Such a sharp change has already taken place in the way in which management purchases (leases) computation equipment. Based upon surprisingly meager information, hundreds of millions of dollars have been conmfitted for the acquisition of new equipment for time-sharing applications. Yet no one has ever seen one of these time-sharing systems in operation, much less learned of its performance characteristics. Past experience is of little help in the evaluation of these systems, and neither manufacturer nor purchaser can realistically answer such simple questions as "How many terminals can a particular system configuration support?" As many managers are now coming to realize, the acquisition of one of these new time-sharin...

An Analysis of Computer Center Operation

“…This time depends on the number of cards in the deck, the type of instructions, and the speed of the card reader. The mean service rate for a category i job is estimated according to pi = nilti (2) 4~sually 8:00 -11:OO am, 1:00 -2:00 pm, and 3:00 where ti is the time it takes to read ni jobs of category i. These estimates are presented in Table 2.…”

Programming systems and languages 1965-1975

“…The early performance was predicted with some accuracy in a simulation study by Nielsen [26], who was then a graduate student at Stanford University.…”

An approach to the optimum design of computer graphics systems

“…A relatively simple model has been chosen, so that many points in its parameter space can be examined. An excellent example of a converse choice is the work done by Nielsen [15], which uses a highly detailed model of the IBM 360/67 and its timesharing software to predict user response time and system loading. Because of the model's detail and its resulting long analysis time, only a single, rather haphazard optimization was possible.…”