Time-and-Traffic-aware collaborative task offloading with service caching-replacement in cloud-assisted mobile edge computing

Abstract The rapid growth of Internet of Things (IoT) applications has increased the demand for ultra-low-latency and energy-efficient computing. While Mobile Edge Computing (MEC) addresses these demands by shifting computation from the centralized cloud to edge servers, its limited resources pose a major challenge. In particular, making optimal decisions for service caching and task offloading under dynamic network conditions and energy constraints remains a critical issue. Efficient caching is essential for latency-sensitive IoT tasks, yet only a subset of services can be stored at MEC-enabled base stations (BSs) due to storage limitations. This paper proposes a Cloud-assisted MEC framework that jointly optimizes service caching, service replacement, and task offloading to enhance long-term system performance. A two-phase solution is developed: first, an Irregular Cellular Learning Automata (ICLA)-based algorithm classifies traffic patterns and timescales, and a Distributed Deep Reinforcement Learning (DDRL) algorithm performs adaptive, decentralized task offloading. To address caching constraints, a dynamic 0–1 knapsack approach selects services based on popularity, while a Q-learning-based policy handles service replacement. Simulation results validate the framework’s effectiveness, showing significant reductions in service latency and energy usage, with improved scalability and adaptability over traditional centralized approaches. The proposed method offers a robust and practical solution for next-generation MEC systems supporting real-time IoT services.