Cache-assisted task offloading strategy based on multi-agent deep reinforcement learning

Abstract

In the context of 5G architecture, mobile edge computing (MEC) has emerged as a promising computing paradigm within this framework. It involves deploying servers at the edge of the network with computing and storage resources to meet the low-latency and high-energy requirements of emerging applications. While many studies have focused on task offloading decisions in MEC, the performance improvement brought by content caching has often been neglected. In this paper, the joint optimization problem of task offloading and content caching in a multi-user, multi-server MEC system is investigated. An offloading method based on multi-agent deep reinforcement learning is proposed, which integrates self-attention with multi-agent deep deterministic policy gradient (MADDPG) to determine task offloading policies. To further enhance the performance of offloading tasks, a dynamic cache decision optimization algorithm (DCDOA) is developed for more effective content caching decisions. The simulation results demonstrate that the proposed method outperforms other baseline algorithms in different scenarios. Specifically, the proposed method reduces the long-term average user cost by 45.4%, 22.2%, 10.6%, and 8%, respectively, compared to the all-cloud execution method, the random offloading method, the DDPG method, and the MADDPG method.