Performance Analysis and Optimization via Simulation

Sammanfattning: A large amount of scientific work has been carried out to enhance the methods used for design and control of technical systems. For many of these systems, the main objective is to utilize the limited resources in an optimal way. Different simulation techniques have, in many cases, been successfully applied to the analysis. When working with technical systems, it is important to have a common notation to define the mathematical models used. Such a notation is presented and applied for describing some realistic models. Load sharing is the process of distributing the total load among the processors in a system. Improvements of different load sharing algorithms in distributed systems are proposed and evaluated by different simulation techniques. The mathematical model formalism, which is a very general and dense formalism for defining and describing mathematical models, is also presented. The formalism, which has been extensively used herein, is based on a few simple and well-known concepts, that are exact, easy to use, learn and understand. It provides people, working with mathematical models, with a ``standardized'' language. The Adaptive Cross Validation method is a very general method to increase the effectiveness of simulation as an optimization tool for all types of dynamical systems. A number of simulation models are executed in parallel with a real system or a model of a system. The simulation results are analyzed to be able to change the system parameters in real-time. This approach makes it possible to use more realistic queueing models without making any assumptions about, for example, the interarrival times and service time distributions involved. The method has successfully been applied to real-time control and design of several types of communication systems. Finally, another novel simulation method for modular optimization, which assumes stationary stochastic processes, is presented. The method is used for dimensioning a circuit switched network with arbitrarily modular link.