Effects of Storage Contention on System Performance

paper presents a mathematical model to measure the amount by which a computer's speed i s reduced when it time-shares storage with other computers and I/() channels. The method can be applied to any number of processors and/or channels and storage units, although the complexity of the solution does increase rapidly as the number of processors increases. Explicit formulas and numerical results are given for several special cases. The results of a simulation of u shared-memory multiprocessor are presented, showing how closely the mathematical model fits the operation of a simulated system. A central processing unit may be required to share main storage units with other central processing units as well as wit'h input/out-put channels. The effect on speed of a reference processor contending for use of a storage unit, can be determined to a reasonable degree of accuracy by use of a stretching factor. The time needed to execute a program without contention is multiplied by the stretching factor to determine execution time in the presence of contention. A mathematical model, consist'ing of a number of interacting processors and shared main storage units, allows determination of the stretching factor. This paper describes the mathematical model, including the simplifying assumptions made, and considers applicat'ion of the method to several processor-storage unit combinations. It then compares results with those obtained by simulation of a shared-storage multiprocessor. The mathematical model In order to make the mathematical analysis tractable, certain assumptions and simplifications are made. However, a restrictive assumption does not reduce the scope of application. Indeed, one must always be prepared to optimally fit a model to its image. The following conventions apply in this paper: