Q- Learning Based Adaptive Multipath Routing Algorithm for Data Centre Networks

This paper presents a Q-Iearning-based adaptive multipath routing algorithm designed specifically for data center networks. By leveraging the discrete nature of Q-Iearning, the proposed method optimizes path selection and sub-flow allocation based on real-time network conditions. The algorithm dynamically adjusts routing strategies in response to changes in network states, enhancing performance metrics such as latency, throughput, and task transmission success rate. Extensive experimental results demonstrate the effectiveness of the Q- learning approach in improving these metrics, offering a simpler yet robust alternative to deep reinforcement learning methods like Deep Deterministic Policy Gradient (DDPG). The proposed algorithm addresses the limitations of traditional multipath routing algorithms, such as Equal Cost Multipath (ECMP) and Weighted Cost Multipath (WCMP), which rely on static link states and fail to adapt to real-time network changes. This work provides a promising solution for optimizing network performance in dynamic environments, making it suitable for modern cloud computing and storage infrastructures that require exceptional performance, minimal latency, and significant throughput.