Dynamic Service Deployment Strategy using Reinforcement Learning in Edge Networks

Edge computing has lately appeared as a practical entrance to intensify the capabilities of the industrial networks by transferring the computation data to the edge of the networks. However, the challenge here is to locate processing devices and provide services through the shortest path by making a proper trade-off between latency and energy of the processing. These demands can be solved using complex optimization problems within a confined deadline, which is barely feasible with traditional statistical techniques. Thus, we propose a heuristic-based task controlling mechanism that makes service deployment decisions by eliminating the need to solve combinatorial optimization problems. Moreover, to achieve the least possible delay and find the best possible path towards the goal, we adopt Graph Reinforcement Learning (GRL) technique. Extensive numerical results illustrate that our proposed strategy generates near-optimal solutions within the given time-bound.