A distributed deep learning-driven edge caching strategy for industrial IoT networks

Abstract

The Industrial Internet-of-Things (IIoT) refers to the use of interconnected networks of industrial-grade devices to enhance productivities and improve the efficiency of industrial processes. IIoT networks have low tolerance for delay and require timely wireless content access. As such, this study aims to investigate the use of an edge computing model, Multi-access Edge Computing (MEC), and a distributed deep learning-driven edge caching strategy to jointly support ultra-reliable, low-latency content access in IIoT networks. Specifically, the proposed distributed framework harnesses on the computing power of edge servers to run deep learning models concurrently. Using simulated network traffic data, the distributed deep learning-driven edge caching strategy was evaluated based on two key performance indicators, cache hit rate and latency. Simulation and real-time results show that the proposed strategy is able to attain 5-15% higher cache hit rates and 10-22% lower latencies compared to traditional benchmark frameworks, including least recently used, and least frequently used.