EMDTORA: Energy-Aware Multi-User Dependent Task Offloading and Resource Allocation in MEC Using Graph-Enabled DRL

The dawn of the 5G/6G networking era has led to the widespread adoption of Multi-Access Edge Computing (MEC), a paradigm shift that brings computational resources at the network's edge to enhance device performance and longevity. Additionally, the proliferation of Internet of Things (IoT) has facilitated the development of complex multi-user, multi-edge server environments. In these settings, the interdependence of application tasks makes computational offloading and resource allocation decision-making challenging, but crucial for optimizing energy efficiency. As a response, in this paper, we propose an Energy-Aware Multi-user Dependent Task Offloading and Resource Allocation (EMDTORA) scheme for IoT-MEC infrastructures. First, we formulate an offline, task offloading, multi-objective optimization problem that aims to minimize the user devices' energy consumption and experienced delay under given constraints. Given the NP-hardness of the problem, we devise an online framework that combines a Graph Attention Networks (GAT)-based mechanism, which captures the in-depth dependency structure of the applications, with an actor-critic off-policy Deep Reinforcement Learning (DRL) algorithm to approximate the optimal solution. Through extensive simulation we highlight the potency of the proposed scheme, as EMDTORA outperforms various baselines by successfully balancing the trade-off between energy consumption and delay minimization, under dynamic network conditions.