Throughput Analysis of Some Idealized Input, Output, and Compute Overlap Configurations
ACM Computing Surveys (CSUR)
The simulation of time sharing systems
Communications of the ACM
Operating Systems Theory
A probabilistic framework for system performance evaluation
Proceedings of the SIGOPS workshop on System performance evaluation
An analytic model of multiprogrammed computing
AFIPS '69 (Spring) Proceedings of the May 14-16, 1969, spring joint computer conference
| Hi-index | 0.00 |
The analytic models used to evaluate the performance of multi-access computer systems are generally queueing models. Their ability to represent, often adequately, these systems is due both to the congestion situation created by the processes competing for a system's resources and to the variability of the workload, whose parameters can be much more concisely characterized as stochastic variables than by some deterministic representation. Thus, deterministic models have been only applied in extremely simple, first-approximation studies (see, for example, Hellerman and Smith and Fenichel and Grossman), or in the analysis of some resource management algorithms (e.g., the problem of optimal schedules discussed in Chapter 3 of Coffman and Denning). This paper presents a non-queueing model of a multiprogrammed interactive system. The model is substantially more sophisticated and more accurate than the deterministic models previously proposed, but retains the characteristics, common to all deterministic models, of greater mathematical simplicity and better understandability with respect to queueing models of comparable accuracy. Even though the equations of the model derived in the paper for an XDS 940 installation cannot probably be applied to other systems, the authors feel that the approach can be used for most systems to obtain quick and reasonably accurate estimates of some performance indices.