A Research On Adaptive Active Queue Management Algorithm

Network congestion refers to the performance degradation caused by the scarcity resources in intermediate nodes when the number of packets transmitted in the packet switched network is too large.Recently,with the development and popularization of cloud computing,mobile Internet and Internet of Things,communication network is becoming more and more complex,which greatly increases the difficulty of network resource scheduling and management.At the same time,the emergency of applications such as e-commerce,telemedicine,distributed computing,instant messaging,Voice over Internet Protocol and network video has increased people's dependence on the network,and put forward many new requirements for the reliability,security and availability of the network.In order to provide users with reliable network quality of service and avoid data loss,delay increase,throughput degradation or even congestion collapse cased by network congestion,it is necessary to design reasonable algorithms for effective management or scheduling of network resources.Active queue management(AQM)algorithm is an important part of network congestion control.It has unique advantages in solving “bufferbloat” problem,improving network service quality and maintaining bandwidth fairness among traffic flows.Considering the time-variety,heterogeneity and complexity of the network,this paper will analyze the system characteristics and design AQM algorithm from the perspective of adaptive control.Difficult issues involved in the algorithm design such as system modeling,time delay processing,parameter identification and fairness guarantee,are thoroughly studied,and the following results are acquired:(1)An information compression model for TCP/AQM system from the perspective of router is established by real-time evaluating the average effect of packet receiving and dropping.The issue of selecting congestion indicator variables is also discussed in the process of modeling.Existing TCP/AQM models usually lack sufficient consideration of heterogenous round-trip times,uncertain endpoint mechanisms,and time-varying network conditions.The assumptions made about the system are also hard to meet in reality and the deduced model contain many unknown parameters.Therefore,exiting models are more suitable for performance analysis rather than controller design.The proposed information compression model can overcome or alleviate these shortcomings,be independent of specific endpoint mechanisms,and track the time-varying network environments in time if accompanied with a customized parameter identification algorithm.(2)TCP/AQM system is essentially a multi-delay coupled system.The round-trip time of each loop is diverse,whose distribution is often far large than the input regulating period.Thus it becomes difficult to accurately access the control effect.Motivated by the generalized predictive adaptive control,the information compression model is appropriately transformed and extended on the time scale to evaluate the input adjustment process on a larger time scale.Compared to existing algorithms(e.g.Co Del),the derived AQM algorithm effectively improves the performance in long delay network conditions.(3)Information compression model is a dynamic approximation model of the original first principle model.Although its form is more concise,its parameters vary widely,which demands better identification algorithms.Through the analysis of classical parameter identification algorithms,it is found that the identification algorithms can be reconsidered from a new perspective of model error allocation.Using this novel analytical framework and considering TCP/AQM system's characteristics and control requirements,a new identification algorithm based on model effectiveness evaluation is designed.The algorithm can effectively deal with the negative impact caused by delay uncertainty,meeting the design requirements of AQM algorithms.(4)Summarizing the gains and losses of previous algorithms,a fairness-driven AQM algorithm integrating protection and monitoring is designed in two steps.To achieve bandwidth fairness as much as possible at small cost is the design purpose and major issue of fairness-driven AQM algorithms.How to effectively identify and punish bandwidth preemption flows is the focus of previous research,however no results satisfactory enough have been achieved.First,the designed algorithm provides necessary protection for the flows already experiencing packet lost.While protecting it,the algorithm also collects some status information for the flows,so as to judge whether its behaviour is legal or not.If it is deemed to be an illegal data flow the protection will be removed,and the algorithm will transfer into monitoring mode and impose necessary penalties on it.The simulation results show that the new algorithm can achieve ideal bandwidth fairness by recording necessary status information for data flows,and achieve significant performance improvements compared to algorithms like g CHOKe.Finally,the paper is summarized and future research is prospected.