Lookahead scheduling in a real-time context: Models, algorithms, and analysis

Our research considers job scheduling, a special type of resource assignment problem. For example, at a cross-docking facility trucks must be assigned to doors where they will be unloaded. The cargo on each truck has various destinations within the facility, and the unloading time for a truck is dependent on the distance from the assigned door to these destinations. The goal is to assign the trucks to doors while minimizing the amount of time to unload all trucks.;We study scheduling algorithms for problems like the cross-docking example that are different from traditional algorithms in two ways. First, we utilize real-time, where the algorithm executes at the same time as when the jobs are handled. Because the time used by the algorithm to make decisions cannot be used to complete a job, these decisions must be made quickly Second, our algorithms utilize lookahead, or partial knowledge of jobs that will arrive in the future.;The three goals of this research were to demonstrate that lookahead algorithms can be implemented effectively in a real-time context, to measure the amount of improvement gained by utilizing lookahead, and to explore the conditions in which lookahead is beneficial.;We present a model suitable for representing problems that include lookahead in a real-time context. Using this model, we develop lookahead algorithms for two important job scheduling systems and argue that these algorithms make decisions efficiently. We then study the performance of lookahead algorithms using mathematical analysis and simulation.;Our results provide a detailed picture of the behavior of lookahead algorithms in a real-time context. Our analytical study shows that lookahead algorithms produce schedules that are significantly better than those without lookahead. We also found that utilizing Lookahead-1, or knowledge of the next arriving job, produces substantial improvement while requiring the least effort to design. When more lookahead information is used, the solutions are better, but the amount of improvement is not significantly larger than a Lookahead-1 algorithm. Further, algorithms utilizing more lookahead are more complex to design, implement, and analyze. We conclude that Lookahead-1 algorithms are the best balance between improvement and design effort