Abstract

The paper develops a class of analytical models for large computer time-sharing systems. The models developed allow the various resources of the time-sharing system, such as central processors, paging devices, and mass storage equipment to be represented as independent queueing centers. The success of the method depends on developing the queueing network so that it may be treated as a time-continuous markov process. Consideration is given to a number of proposed measures for system load and system performance. These measures are examined in the light of the model. In addition, concepts of system saturation and system balance are introduced and defined in terms of the model. As an example of the use of network models, a model is developed for the University of Michigan Terminal System (MTS). Statistics available from that system are used to develop and validate the model. Finally, possible changes in that system's structure, in light of the insights gained from the model, are explored. (Author)