A Sub-flow Adaptive Multipath Routing Algorithm for Data Centre Network

Abstract

Abstract In response to the fast and intensive traffic changes and the presence of redundant links in the network topology in data centre networks, multipath routing has become the dominant approach. Existing multipath routing algorithms are still inadequate in terms of adapting to dynamic changes in the network state quickly and the cooperation of path selection and sub-flow assignment. Therefore, this paper proposes a sub-flow adaptive multipath routing algorithm for data centre networks (SAMP). The deep deterministic policy gradient (DDPG) algorithm is introduced. DDPG combines deep learning (DL) and reinforcement learning (RL) to implement different network state changes quickly, especially in the process of topology, to achieve dynamic migration of the optimised decisions that have been learned. An adaptive multipath routing model for subflows is established to accomplish collaborative scheduling of path selection and subflow assignment based on the real-time state of the network. The experimental results show that the algorithm can be well adapted to the data centre network, and when compared with several traditional methods, it reduces the latency by an average of 31.3%, improves the task transmission success rate by an average of 14.9%, and increases the throughput by 37.1%.

Keywords

• Multipath routing
• Reinforcement learning
• Data center network
• Traffic schedule
• Software defined network