End-Edge Collaborative Optimization of Microservice Caching in D2D-Assisted Network

Employing the caching resources of end users via Device-to-Device (D2D) communication to assist the edge server in microservice caching is promising to further alleviate the network congestion of the Internet of Things (IoT). However, significant extra energy consumption prevents the caching system from maximizing cache utility if all end users cache simultaneously. In this paper, we propose two novel end-edge collaborative microservice caching algorithms in D2D-assisted networks. First, we construct a D2D caching sharing link graph from the aspects of physical and social attributes of end users and introduce the Entropy-based Partitioning Around Medoid (EPAM) algorithm to identify critical users. Second, to address the challenges posed by unknown time-varying user preferences, we model the end-edge collaborative caching problem as a Multi-Agent Multi-Armed Bandit (MAMAB) problem, thus developing two caching decision schemes, i.e, Edge-Centric Scheme (ECS) and User-Centric Scheme (UCS), to accommodate different decision sequences. The simulation results show that the EPAM-ECS and EPAM-UCS have at least 29.2% and 39.3% improvement compared with other baseline algorithms.