Stochastic Local Search for Multiprocessor Scheduling for Minimum Total Tardiness

The multi-processor total tardiness problem (MPTTP) is an -hard scheduling problem, in which the goal is to minimise the tardiness of a set of jobs that are processed on a number of processors. Exact algorithms like branch and bound have proven to be impractical for the MPTTP, leaving stochastic local search (SLS) algorithms as the main alternative to find high-quality schedules. Among the available SLS techniques, iterated local search (ILS) has been shown to be an effective algorithm for the single processor case. Here we extend this technique to the multi-processor case, but our computational results indicate that ILS performance is not fully satisfying. To enhance ILS performance, we consider the use of population-based ILS extensions. Our final experimental results show that the usage of a population of search trajectories yields a more robust algorithm capable of finding best known solutions to difficult instances more reliably than a single ILS trajectory.