Autonomous Cache Resource Slicing and Content Placement at Virtualized Mobile Edge Network

Abstract

Nowadays, the unprecedented growth of mobile traffic is only rivaled by the demands for a better quality of service and faster data delivery. Caching the content closer to the end users is one solution to cope with these demands, but when service providers share a common physical infrastructure, managing the limited cache resources becomes more challenging. With this increased complexity, the deep reinforcement learning (DRL) presents itself as an optimal solution. In this work, we propose a dynamic DRL-based management framework for virtual cache slicing and customized content placement. The virtualization problem is formulated as a Markov decision process, and it utilizes a deep Q-network to autonomously make network-wide slicing decisions and optimize the cache resource allocation. These global decisions are then mapped to the small cell stations with a dynamic resource provisioning algorithm. Lastly, a customized content placement algorithm is used to find the optimal content placement policy for each service provider. The content placement problem is first formulated as a convex optimization problem and is then solved with a distributed alternating direction method of multipliers (ADMM). Finally, the simulation results are presented to show the effectiveness of the proposed solution.

Keywords

• Content placement
• deep reinforcement learning
• deep Q-networks
• network virtualization
• resource provisioning
• resource slicing