Study Of Network Congestion Control In Information System

With the continuous expansion of computing network scale and increase of application type,congestion control is becoming more and more importance in ensuring the operation of the network and quality of service,which is mainly composed of TCP(Transmission Control Protocol)and AQM(Active Queue Managment).AQM mechanisms which deployed in router can monitor network status proactively,and have advantages such as take the initiative to avoid network congestion,improve link utilization and the dropout rate.But studies show that the AQM algorithm has difficulties of parameter configuration,queues performance instability,lack of low link utilization when facing network scene such as dynamic mutation and mixed flow,and it attracted the attention of academic.The thesis is supported by ministerial research"Study of network control in ship information system",research of AQM algorithm and its related problems.The T-S(Takagi-Sugeno)fuzzy model has characteristics of precise mathematical model requireless and good approximation ability of nonlinear system.The T-S fuzzy model of TCP/IP network congestion control system is established with equivalent the network system affected by the link parameter perturbations as a state-space model with uncertainties.Lyapunov stability theory and linear matrix inequalities(Linear Matrix Inequlity,LMI)is ued to give the sufficient condition of state feedback loop system is asymptotically stable and the design method of controller.The design method of fuzzy guaranteed cost control law is given,so as to achieve the specified performance optimization design.The simulation results show that the designed controller can guarantee the queue length of router track target stablely and avoid network congestion occurs when the number of TCP connections and the round-trip delay change with uncertainties.Research show that the model of network congestion has non-linear characteristic,and the network parameters changes in operation,so it required to consider changes of model parameters.In such cases,the active queue management algorithm based on predictive function control is designed which use predictive model to predict future dynamic of network and improve the performance of the control system to a certain extent,so that the queue length can track target stablely.The solving of network predict control with performance constraints is changed into Lagurange multiplier algorithm which avoid solve the problem of the QP in router directly and computing resources is saved.The algorithm strong robustness when the parameter perturbations,dynamic burst stream and non-response stream interference occur in network.Predict sliding queue management strategy is proposed for the TCP/IP network model,the algorithm use predictive models to predict future dynamic network sliding surface,uses sliding mode control with good transient performance and robustness characteristics,optimizes the packet loss rate to ensure router queue length each expectation quickly and smoothly.Study the influence of AQM system stability when time-delay changing in network,linear matrix inequality and Lyapunov-Krasovskii theory has been used to get sufficient conditions which proof the stability of network system does not rely on time-delay.Simulation results show the sliding mode prediction algorithm has good control performance of network with large delay,overcome the adverse effects for system caused by uncertainties such as network burst stream and round-trip delay,avoid network congestion occur effectively.Congestion control method based on complex adaptive model difficult to meet the needs of high-speed networks in speed,easy handling and other aspects,but fixed simplified mathematical model will lead to loss of control precision and performance.Grey prediction modeling method is introduced to deal with uncertain variables so that the model has the ability to compensate for the poor robustness caused by fixed model,robust adaptive controller is added in the sliding mode controller.The online learn method can adjust the control parameters adaptively in order to ensure stability of network system in the vicinity of the sliding surface,improve the characteristics of response speed and steady-state.Simulation results in complex multi-network bottleneck scenario show that the control strategy can inhibit the network disturbance effectively,it has strong robustness for hybrid heterogeneous networks.Describes the large scale network information system which author participate in design.Discusses the basic function and structure of the network system,describe the hardware architecture and monitoring software system design principles simulation platform,focusing on a combination of hardware and software semi-physical simulation platform,as well as rudder propeller cooperative control and other properties of experimental projects,provide a reference for the future formation of a ship information network system.The project has an important value of work.Starting from various angles such as theoretical significance and application prospects,thesis have research and demonstration of active queue management algorithm proposed.Gives a detailed parameter configuration program and adjust rules.Moreover,conduct a series of simulation experiments such as single and multi-bottleneck network scenario,given the results of a detailed analysis of algorithm performance comparison.Simulation results show good performance of several AQM control algorithms presented herein and different characteristics.