Abstract:
In this study we have introduced a mathematical foundation and a framework for the modeling and analysis of agent based distributed scheduling systems. We defined the basic structure of an object oriented agent based system, and the issues in distributed systems. In order to explore the agent based structures deeply, a state based modeling approach is used. We combine a set of decision variables or control variables under the name of control policy. The control policy is comprised of the decision process (DP), response policy (RP), and agent stability during the decision process. To test the impact of various control policies to the performance of the system we set up an experiment with different problem settings based on due date tightness, the frequency of job arrivals, and processing time distribution. In this study we have been able to show that the solution procedure by itself does not guarantee a good overall performance in the system when the control policy is not set up correctly. Also we have been able to show the performance of these control policies change from one problem setting environment to another. So it is not possible to say that one magic solution procedure will be able to solve all the problem sets. It has been showed that introducing the decision time to the decision process leads to significantly poorer results compared to instantaneous policies, and even a responsive random dispatch policy may perform well better compared to a takes time control policy equipped with an optimum dispatch algorithm.
