A Reconfigurable Optimizing Scheduler

We have created a framework that provides a way to represent a wide range of scheduling and assignment problems across many domains. We have also created an optimizing scheduler that can, without modification, solve any problem represented using this framework. The three components of a problem representation are the metadata, the data, and the scheduling semantics. The scheduler performs the optimization using an order-based genetic algorithm to feed different task orderings to a greedy schedule/assignment builder. The scheduler obeys the hard and soft constraints specified in the scheduling semantics. We have applied this reconfigurable scheduler to a variety of scheduling and assignment problems including the job shop, traveling salesman, vehicle routing, and generalized assignment problems. The results demonstrate that the optimizer can provide not only easy reconfigurability but also competitive performance.