Study On Situation Awareness Of Inter-network Performance In Internet

Performance situation is the synthesis of network performance,and its awareness refers to the comprehensive and continuous monitoring of the performance of various parts in networks so that users and operators can have a comprehensive and accurate understanding of network services as well as the overall performance of networks.With the rapid development of computer networks,they have offered all kinds of services that have penetrated into people’s lives.Network management,especially performance management,is increasingly important because it is the key to providing stable and efficient network services.As a vital task of performance management,performance situation awareness possesses tremendous research significance.The measurement methods used in performance situation awareness pursue the topological coverage of monitoring,which can tolerate low availability,but require low measurement costs.However,the main measurement methods currently applied in performance management are diagnosis-oriented,which come with high costs because the measurement need to be targeted and timely,in pursuit of the availability of data.Therefore,those current measurement methods are not feasible for performance situation awareness.Due to the unavailability of applying current measurement methods to network performance situation awareness,this paper is aimed to propose a method to sense the performance situation for large-scale networks.With the application of passive measurement technology,it needs much fewer deployment and maintenance costs than the active measurement does.It achieves a high coverage of monitoring and lower measurement costs by taking the sampled flow data as data source,which,at present,are generally supported and efficiently collected by a variety of backbone network routers.This paper includes three parts:the first two parts belong to situation perception in situation awareness,and the last part presents a method for situation projection.The first part of this paper is about the RTT estimation methods based on sampled flow data.Two RTT estimation methods are proposed for AIMD TCP flow data and non-AIMD TCP flow data,respectively.Based on the analysis of the transmission features of TCP bulk flows when the socket buffer is bigger than and smaller than BDP,each method establishes two estimation models for these two situations,and comes with an efficient method to distinguish these two situations.And these methods are also suitable to sampled flow data because while estimating RTT,they only apply the duration time and total number of packets.Consequently,the cost of collecting and storing source data for performance situation awareness can be largely reduced.The experiments indicate that the proposed methods can achieve estimation results approximate to those attained by full packet trace;therefore,our method can meet the requirements of performance situation awareness.In the second part of this paper,a network tomography based on passive measurement is introduced.In spite of the advantageous passive measured data,two problems need to be solved in order to apply network tomography:1)the systems of linear equations related to network tomography may have infinite solutions because the routes through which the traffic passes are determined by the routing table,thus cannot be modified during the measurement process;2)network tomography technology may not be feasible at all due to missing data caused by the possible failure to collect the relevant traffic or probable measurement errors.This paper presents DDSP algorithms and employs compressive sensing method to solve the above problems respectively,which makes it possible to take the passively-collected,low-cost RTT data as input to estimate the delays of inner paths.The experiments demonstrate that the paths with moderately short lengths generated by DDSP algorithms are able to cover the vast majority of links;moreover,our proposed method can achieve high accuracy in the absence of some data and in the presence of many outliers.The third part of the thesis introduces the method to locate network performance anomalies with WPM-SAT.The prior works are either with application limitations or not suitable to performance anomaly localization.Based on the abnormal values of the path delays calculated by the network tomography introduced above,a method is proposed to locate the network elements with potential performance anomalies:modeling this problem with WPM-SAT and getting the result by solving it.Since the WPM-SAT problem is NP-hard,the standard algorithm may take too much time in some cases;thus,an approximation algorithm with polynomial time complexity is also presented,as the supplement to the standard algorithm.The method does not need the cooperation of inner nodes and has no requirement of network topology and measurement paths.Furthermore,as the experiments show,with similar running time,the proposed method based on two different WPM-SAT algorithms can obtain higher accuracy than the previous localization methods.As a result,the proposed method is more applicable for performance anomaly localization.This thesis is the primitive exploration of performance situation awareness.Our methods,at a much lower cost,can help network operators achieve results similar to these obtained by performance diagnostic tools.With the additional availability to locate potential performance anomalies,our methods can achieve performance situation awareness for large-scale networks.Therefore,the thesis meets the requirement of inter-network performance awareness and can inspire follow-up studies.