Decomposition and learning for a real time task allocation problem
Résumé
We present a cooperation technique using an accurate management of nogoods to solve a hard real-time problem which consists in assigning periodic tasks to processors in the context of fixed priorities preemptive scheduling. The problem is to be solved off-line and our solving strategy is related to the logic based Benders decomposition. A master problem is solved using constraint programming whereas subproblems are solved with schedulability analysis techniques coupled with an ad hoc nogood computation algorithm. Constraints and nogoods are learnt during the process and play a role close to Benders cuts.