Deep Reinforcement Learning Based Approach for Balanced Data Deduplication in Mobile Edge Computing

Mobile edge computing (MEC) migrates computing, storage, and network resources to the network edge, providing users with low-latency data access. As a crucial part of MEC, edge storage systems (ESS) offer storage capabilities near data sources or users. However, the limited server storage resources present challenges for ESS in optimizing storage utilization efficiency. This paper investigates a balanced data deduplication problem in ESS involving multiple collaborative servers and different tasks while comprehensively considering deduplication rate, storage balance, and data storage efficiency. To maximize overall system performance and overcome the limitations of traditional methods that rely on static rules, we formulate the problem as a Markov decision process (MDP) and propose a deep deterministic policy gradient (DDPG)-based approach to find the optimal deduplication solution. We employ a dual-network structure combined with a learning rate scheduler and implement prioritized experience replay to enhance training stability. Simulation results demonstrate that our approach outperforms the baselines across various parameter settings.