Analysis of the impact of decision time on the system performance in distributed systems

In this study, we investigate the effect of the time it takes to generate a schedule on the performance of a stochastic dynamic scheduling system. To isolate the impact of the decision time on the system performance, we devise a single machine stochastic scheduling environment where the performance of the system is measured by average earliness - tardiness cost. Our study is composed of two phases. In the first phase, we construct a centralized scheduling system. We explicitly model the decision time. We test the trade off between spending more time for the scheduling process by employing more sophisticated scheduling algorithms and using simple fast heuristic algorithm. In the second phase, we construct a distributed scheduling system. We test the trade off between spending more time by including detailed global information to achieve global optimality under a centralized control structure and using timely accessible local information under distributed control. We simulated the system under various scheduling environments controlled by due date tightness, urgent job ratios, operation time variability and utilization using different centralized control polices and distributed control policies. Our experiments show that under certain shop conditions and control policies, the shop may operate more efficiently if a simple fast heuristic is used instead of a slow optimum algorithm to solve the scheduling problems. We have been able to also show that, again under some specific operating conditions, the dynamic production system will run more efficiently when we use fast distributed schedulers instead of a relatively slow centralized scheduler.