Exploring the use of Deep Reinforcement Learning to allocate tasks in Critical Adaptive Distributed Embedded Systems

Abstract

[eng] A Critical Adaptive Distributed Embedded System (CADES) is a group of interconnected nodes that must carry out a set of tasks to achieve a common goal, while fulfilling several requirements associated to their critical (e.g. hard realtime requirements) and adaptive nature. In these systems, a key challenge is to solve, in a timely manner, the combinatorial optimization problem involved in finding the best way to allocate the tasks to the available nodes (i.e. the tasks allocation) taking into account aspects such as the computational costs of the tasks and the computational capacity of the nodes. This problem is not trivial and there is no known polynomial time algorithm to find the optimal solution. Several studies have proposed Deep Reinforcement Learning (DRL) approaches to solve combinatorial optimization problems and, in this work, we explore the application of such approaches to the tasks allocation problem in CADESs. We first discuss the potential advantages of using a DRL-based approach over several heuristic-based approaches to allocate tasks in CADESs, and we then demonstrate how a DRLbased approach can achieve similar results to the best performing heuristic in terms of optimality of the allocation, while requiring less time to generate such allocation.