Q-Learning Algorithm for Joint Computation Offloading and Resource Allocation in Edge Cloud

The advent of 5G technology along with the high proliferation of mobile devices entail an explosion of mobile traffic. Due to their resource-limitation constraint, mobile devices resort to connect to Cloud servers so as to offload computational tasks and improve, hence, resource usage. Unfortunately, the conventional Mobile Cloud Computing (MCC) solution involves high transmission latency. Recently Mobile Edge Computing (MEC) is envisioned as a promising technique for enhancing the computation capacities of mobiles and reducing latency. The key insight of MEC is to push mobile computing and storage to the network edge (i.e., base stations and access points). The main challenge of MEC solution is to find an efficient assignment of tasks with local or edge devices, while minimizing energy consumption and latency. In this paper, we propose a new joint task assignment and resource allocation approach in a multi-user WiFi-based MEC architecture. The main novelty of our work is that optimal offloading decision is jointly performed with the radio resource allocation. The objective of our scheme is to minimize the energy consumption on the mobile terminal side under the application latency constraint. To do so, we first formulate the problem as a new online Reinforcement Learning problem while considering both delay and device computation constraints. Then, we propose a new strategy based on a Q-Learning algorithm, named QL-Joint Task Assignment and Resource Allocation (QL-JTAR) to solve it. Based on extensive simulations conducted in NS3 simulator and using real input traces, we show that our approach outperforms the related prominent strategies in terms of energy consumption and delay, while ensuring near-optimal solution.