Optimized Multipath Transmission System For Data Center Networks

With the continuous increasement of the scale of modern data centers,various emerging technologies have also been applied to data center networks to improve throughput efficiency and reduce latency,and MPTCP(Multi-Path Transport Layer Protocol)technology is one of them.MPTCP is a type of multi-path transmission technology.Its core idea is to divide the congestion window and establish multiple subflows accordingly when the transmission connection is set up;Packets will be sent through different subflows,and at the receiver they will be collected/ordered.MPTCP has been widely used in the field of computer networks and has achieved significant effects of improving bandwidth,enhancing stability and redundancy.Also,academia as well as industry expects this technology to be deployed in data center networks,so that significant performance improvements in large-scale data transmission can be achieved.However,researchers have found that MPTCP did not perform as expected when deployed in the data center.Theoretically and experimentally,MPTCP has the following two problems when deployed: On the one hand,as a transport layer protocol,MPTCP does not care about the specific physical path direction of the sub-stream,which leads to the difference when it is deployed in the data center network.Subflows may use a large number of repeated paths,which cannot effectively increase the transmission bandwidth and will cause resource preemption to other conventional TCP flows and further deteriorate the data center network conditions;in addition,MPTCP’s congestion control protocol restricts the total transmission bandwidth of subflows too strictly,which make it unable to guarantee its own transmission capacity not affected by burst traffic and unable to meet our requirements for setting the protocol to ensure high-priority connection’s transmission.Therefore,how to adapt to the characteristics of the data center network environment and give full play to the performance advantages of the MPTCP protocol,is the important problem to be solved in this paper.In order to solve the problems above,this master’s thesis carries out research work from the following three aspects: First,for the problem of poor transmission performance caused by the repetition of subflows’ physical links,this paper proposes a minimum-mark-based subflow physical path detection and congestion avoidance mechanism which uses the switch flow rule to push the path label information into the packet header,and then calculate the subflows’ path repetition degree based on the path information,delete and recreate the high-duplication subflows,so as to encourage the sub-flow to find the low-load link,reducing the subflow path repetition and improving bandwidth;Secondly,in view of the low transmission efficiency of MPTCP caused by the dynamic change of the sub-flow load on the heterogeneous path,this paper proposes a subflow active congestion control mechanism based on link history information;this mechanism actively collects the link history information of each sub-flow to model their transmission capacities,and then a fixed excellent congestion window value is calculated based on the modeling and convex optimization theory so as to ensure the transmission speed of the high-priority flow where MPTCP is deployed,while minimizing the impact on other flows in the network,reducing network traffic jitter and packet loss.Finally,based on the two mechanisms above,this paper designs a multi-path transmission optimization system for data center networks,which implements the two algorithms in the NS-3(Network Simulator 3)network simulation platform and conducts performance experiments;Simulation experiments show that the optimized system proposed in this paper consumes low system resources and can significantly improve the data transmission bandwidth and transmission stability of MPTCP in the data center.To sum up,this paper proposes an MPTCP performance optimization mechanism with low resource consumption and high transmission performance for MPTCP deployment scenarios in data center networks,finally provides an efficient multi-path transmission optimization system for modern data center networks.With the rapid development and popularization of data centers in the world,the research results of this paper can be applied to large-scale data center networks to provide a high-performance and low-cost transmission protocol system for large-scale data transmission within the network,which has important theoretical and application value.