A study of computer time-sharing systems

Computer systems have evolved during the last decade from mono-programmed, batch processed to multi-programmed, remote processed operation. The technique of time-sharing a computer system between many remote users now offers the combined advantage of expanded user facilities and high equipment utilization. This study deals with the definition of various algorithms necessary for time-sharing and the simulation of their actions. Of the several possible approaches to time-sharing, the page on demand method is chosen as the best contemporary approach. The algorithms needed for this method are developed into a model of the generalized time-sharing computer system. The model is then programmed into a simulator of the model's action which receives input of the machine, peripheral, supervisor and program parameters and outputs timing results for individual programs and total timings. Simulations are made of various hypothetical and real machines as follows: (1) Standard computer for comparison, (2) SDS 940, (3) IBM 390/67R, (4) CDC 3300, (5) Standard computer as function of number and size of pages, (6) Standard computer as function of number of pages, (7) Standard computer as function of size of pages, (8) Super computer Results are tabulated and charted for comparison. The results show the relative performances of the commercially available systems to favor the smaller systems with large pages. Pages only have to be larger than a certain minimum which is apparent in the graph of results from variation of this parameter. The variation of parameters to effect number of pages shows a smaller effect on performance, but the combined effect of page number and size is quite noticeable. The simulation of a hypothetical super computer shows the importance of peripheral devices on overall performance. It can be concluded from this study that the technique of time-sharing can yield the desired effect of large facilities to the user and reasonably high efficiency of the system. Although absolute evaluation of results is only meaningful in the light of the program mix which was used, still relative performance lends direction to development. It is also shown that simulation is a useful tool for studying complex systems such as time-sharing of computers