Opti DRL: Integrating Optical Networks With Multi-Objective Optimization And Deep Reinforcement Learning For MEC Resource Allocation

Abstract

The sudden boom of latency-demanding and bandwidth-hungry applications in 5G and beyond has loaded tremendous pressure upon Mobile Edge Computing (MEC) infrastructures. Resource provisioning is important in order to meet the stringent quality-of-service (QoS) demands, while conventional approaches do not perform well with the dynamism and heterogeneity of the MEC setting. Furthermore, current solutions largely ignore the strength of high-speed optical networks as a means for improving MEC performance. We introduce OptiDRL, a new framework that combines optical network infrastructure with deep reinforcement learning (DRL) and multi-objective optimization to enable intelligent and adaptive resource allocation in MEC systems in this paper.

OptiDRL casts MEC resource allocation as a multi-objective decision-making problem, balancing latency, energy usage, and resource utilization. A DRL agent is learned in this context with a well-crafted reward function that balances these objectives. The optical integration provides ultra-low latency and high-throughput communication, further improving system efficiency. We deploy and test OptiDRL on simulation environments mimicking real-world MEC environments.

Experimental results show that OptiDRL outperforms current state-of-the-art benchmark algorithms with a significant latency reduction of up to 35%, resource saving of 25%, and scalability improvement under changing workload scenarios. This paper proves the potential in integrating optical networking with DRL to advance intelligent MEC resource management.