Research On Intelligent Scheduling Technology Of Dynamic Flow For Data Center Network

The digital transformation of society is promoted by the rapid development of network technology and the widespread deployment of network infrastructure.Meanwhile,the construction,intelligent upgrading and transformation of large data centers have been accelerated by the computational transmission and storage of massive data.As a means of utilizing network resources effectively,traffic scheduling technology could optimize network performance and help the network quickly adapt to the business transformation.However,traditional data center architecture and traffic scheduling mechanisms can no longer meet the network’s requirements for low-latency connections and high-quality transmission.Therefore,this thesis mainly carries on research of the dynamic traffic intelligent scheduling mechanism in the data center network based on the software defined network architecture.The specific thesis work and research content are organized as follows.Firstly,this thesis analyzes traffic scheduling mechanisms in current data center network in terms of traffic and resource scheduling problems,such as network load imbalance and low link utilization.Then,the thesis analyzes the application potential of software defined network architecture to the data center network.In addition,related works on key technologies of traffic scheduling in data center networks are introduced.Secondly,according to the surge in traffic in the current data center network and the increasingly complex network requirements,a dynamic flow scheduling mechanism based on deep reinforcement learning is proposed to make full use of network resources.The mechanism utilizes multi-agent reinforcement learning algorithm for traffic scheduling,which could be performed on the pre-calculated path without solving complex mathematical models.Simulation results show that this traffic scheduling mechanism has a significant performance improvement compared with the traffic scheduling mechanism based on single agent reinforcement learning.Besides,the mechanism performs good convergence and robustness under different network architectures,which achieving the purpose of taking advantage of resources in the network.Finally,in view of the serious performance problems and scalability bottlenecks of data centers based on software defined network architecture,a data center network service quality assurance mechanism based on traffic hierarchical detection is proposed to improve network service quality under the premise of ensuring the security of information transmission.Under this mechanism,the controller implements differentiated scheduling strategies for different types of flows according to the results of traffic classification detection to avoid network congestion,thereby improving network performance and ensuring network service quality.Simulation results show that the differentiated flow scheduling mechanism could dynamically schedule flows according to the current network status,which guarantees the good performance of the network.