Multiprocessor system simulation and some measurement of system performance

Abstract

This thesis describes the behavior of the multiprocessor system through simulation. It also describes some simple CPU scheduling policies which are implemented in the system. The job processing sequence of a given uniprocessor system is studied and is used to build a simulation model. The model to be built is considered in macro-level. The modeling is done with GPSS which provides very useful tool for the kind of modelling needed in the thesis. The building of the multiprocessor system model is then represented using the model for uniprocessor system which is first described. The behavior of the I/O system is then studied with the introduction of channel concept to which are connected a number of I/O devices. A number of scheduling policies are studied to find out the improvement in CPU utilization as well as overall performance. Some of the policies considered are FIFO, LIFO, Shortest-service time, Longest-service time, I/O bound and Time-slice round robin. The turnaround times for different I/O configuration are given to demonstrate the kind of I/O system most suitable for the multiprocessor system in terms of performance and resource utilization.