deepV:PPO-driven Task Latency Minimization in Volunteer-supported Edge Computing

Although the emergence of edge computing has greatly enhanced the efficiency of data processing and reduced latency by bringing computation closer to the data source, one significant drawback of edge computing is the lack of processing capabilities of centralized cloud servers. With the rapid development of Internet of Things (IoT) and Artificial Intelligence (AI) techniques, the computational demand on edge servers is increasing exponentially. To address such shortage of resources at edge servers, this paper proposes deep V, a volunteer-supported edge computing framework. We also formulate a long-term multiobjective problem for joint optimization of resource allocation and volunteer employment. First, driven by deep reinforcement learning (DRL), a novel PPO-based algorithm is designed to configure the volunteer to IoT (V2I) device matching as well as uplink bandwidth allocation; Second, an efficient resource allocation and data partitioning algorithm is proposed utilizing the block coordinate decent (BCD) method. Through extensive simulations, we demonstrate the schedulability and scalability improvements of our work over other related algorithms.