Dynamic Resource Allocation of Reinforcement Learning Based on Neural Networks in Software Defined Networks

Abstract

With the rapid development of Software Defined Networking (SDN) technology, how to efficiently and flexibly manage and allocate network resources has become a key challenge. This article proposes the DDPG (Deep Deterministic Policy Gradient) algorithm method, aiming to dynamically optimize resource allocation in SDN. The DDPG algorithm can respond in real-time to changes in network status, automatically adjust resource allocation strategies, and thereby improve network performance and service quality. This study comprehensively evaluated the dynamic resource allocation ability of neural network-based DDPG reinforcement learning algorithm in SDN through four experiments. In the baseline comparison experiment, the network throughput of DDPG reached 95 Mbps. Under different network loads, DDPG still maintains a throughput of 95 Mbps under high load conditions. In the fault recovery capability testing experiment, the recovery time of DDPG is 30 seconds. In the final real-time adjustment capability evaluation, DDPG demonstrated a fast response time of 1.2 seconds, as well as a throughput of up to 80 Mbps and a resource utilization rate of 95% after adjustment. From the experimental data conclusions, it can be seen that the DDPG algorithm provides superior performance and flexible resource management capabilities in SDN environments.