Research On Multipath Transmission Control Strategy Of Heterogeneous Data Center Network

With the acceleration of innovation in the Internet,big data,cloud computing,artificial intelligence and other technologies,the demand for high-quality new digital services has increased significantly.As an important carrier of data information exchange,computing and storage,the scale of data centers continues to expand,and the traffic aggregated within them continues to surge.In addition,the deployment and promotion of network access technology and IPv6 have promoted the increasing popularity of multihomed hosts under data center network architecture.Therefore,data center networks(DCN)urgently need to have multi-path transmission and dynamic scheduling capabilities.Much research is done by deploying multipath technologies to aggregate throughput and maximize data center network resource utilization.Among them,MPTCP(Multipath TCP)has received extensive attention and research from the academic community because of its protocol advantages.However,with the development of data centers,the existing multi-path transmission control schemes face technical challenges such as execution accuracy,scalability,and performance in heterogeneous environments with high dynamics,delay jitter,malicious attacks,and limited resources.In the transmission process,it is easy to face problems such as packet out-of-order arrival,buffer blocking at the receiving end,and reduced transmission robustness,which seriously affects the quality of service.In order to solve the above problems,based on the MPTCP protocol,this paper proposes two optimized data center multipath transmission control schemes to improve data transmission performance and enhance the reliability of heterogeneous data center networks.The main research contents are as follows:(1)In order to alleviate the problem of low data scheduling efficiency caused by the difference in quality of different paths,this paper proposes MPTCP-me Learning based on multi-expert learning for data center multi-path transmission control scheme.Among them,the path quality assessment model enabled by multi-expert learning improves the quality assessment accuracy of paths experiencing sudden changes by aggregating a representative set of throughput estimation models.The data scheduler based on multi-expert path quality assessment can adaptively select high-performance path sets for data transmission,enhancing its robustness in response to sudden changes such as network attacks.(2)The existing multi-path transmission control scheme based on static model is difficult to meet the needs of data center business in terms of complexity and accuracy.Therefore,a novel data center multipath transmission control scheme based on reinforcement learning,l~2-MPTCP,is proposed.Among them,the path quality evaluation model based on forward delay can judge the quality of the path more accurately.The reinforcement learning-driven multipath manager can perceive and learn the network environment in real time,and continuously improve the multipath transmission control strategy when the path fails suddenly,so as to prevent the use of poor performance paths to cause a large number of packets to arrive out of order,so as to achieve efficient and reliable transmission of data center services.(3)The performance of the proposed scheme is verified by the NS2 simulation platform.Experimental results show that when some paths in the data center network experience sudden events such as network attacks,the proposed scheme is superior to the existing heterogeneous data center network multipath transmission control scheme,and can obtain higher network transmission performance and robustness.