Adaptive Reservations for Feedback Control
Résumé
In this paper, we start from an assigned control law with known probability distributions of its execution time. Our goal is to identify an optimised scheduling policy that allows us to strike an acceptable tradeoff between control performance and consumption of computation resource. The methodology we advocate is based on a resource reservation scheduler that allows us to control the budget of CPU time allocated to the task and to construct a dynamic model for the delays introduced by the computation. This model allows us to construct a feedback scheduler (adaptive reservations) that dynamically changes the bandwidth to fulfil the desired application goals minimising the computation resource requirements.