A Parallel and Distributed Data Management Approach for MEC Using the Improved Parameterized Deep Q-Network

With the rapid development of Multi-access Edge Computing (MEC), massive distributed collaborative data are generated in real-time at the edge of the network, which brings great challenges to the parallel and distributed data management for compute-intensive and low-latency application requirements. Traditional offloading strategies often neglect signal interference and the critical interdependence between edge servers. Our work combines the parallel and distributed data management system with the operational requirements of MEC. We employ a Partially Observable Markov Game (POMG) framework alongside an Improved Parameterized Deep Q-Network (I-PDQN) algorithm, specifically designed for complex decision-making scenarios with task offloading and resource allocation in a discrete-continuous hybrid action space. The collaboration between the POMG framework and the I-PDQN algorithm facilitates data management and analysis across multiple edge servers and it addresses the dual challenges of parallel and distributed data management by ensuring data consistency and availability across distributed nodes, while optimizing computational resources to reduce latency and energy consumption. Experimental results illustrate that compared with the start-of-the-art baselines, our approach achieves competitive improvement in energy consumption and time delay.