Research On Quality Of Service Analysis And Guarantee Based On Network Calculus

Compared with the traditional queuing theory, network calculus possesses systemmodel structure and can adopt a flexible manner to express an increasingly complexnetwork morphology and traffic flow characteristics, thus an accurate model of the trafficflow and service unit can be obtained and network calculus has the ability to worst-caseperformance analysis, it is very suitable for the modeling of network QoS.Based on the foregoing background, QoS analysis and guarantee methods of dynamicheterogeneous network are considered in terms of network calculus in this project. Thestudies consist of four parts:1) network modeling for the general topology;2) datatransformation-oriented network service model;3) characteristics-oriented admissioncontrol algorithm, such as self-similar and Multifractal of of traffic flow;4) the routingalgorithm based on the traffic flow and network services with dynamic changes. The fourparts form a whole step by step.Firstly, the three main factors of affecting the stability of the network are analyzed,which include network topology, input network traffic and scheduling protocol ofnetwork service node. The specific details of the impact on network stability are found,which provide a clear target for subsequent study of network stability condition.Concept ofthe hyperlink is exploited for researching network stability conditions, but the hyperlinkdefinition adopts mathematical structure of the vector. So, the basic network calculus theorycan not be used to model the data structure represented by the hyperlink. Therefore, byvirtue of the hyperlink concept, statistical delay bound and statistical backlog bound areinduced based on network calculus, they can be used to express steady metrics of networkelement to investigate network stable conditions with the general topology. By studyingthe partial order relation between data packets or traffic flows and the impact of thispartialorder to delay, the closure operator on the partially ordered set is researched, and theproblem of network stability is changed into super-additive closure problem. Whennetwork is in steady state, general topology networks is modeled through the integrationof the minimum plus algebra and max-algebra convolution, which can solve the problemthat network calculus is only used to model the series topology network.Secondly, because the traditional traffic flow aggregation is not used for the description of the traffic flow aggregation of data transformation, this section gives adefination of traffic flow aggregation which is fit to data transformation. Datatransformation is viewed as a kind of data compression process, and the correspondingprocess is defined from the perspective of functional analysis, service model and arriveprocess model of traffic flow based on this definition are given. According to these twomodels, expression of QoS performance parameter with data transformation capability isdeduced, performance analysis method is introduced to data transformation model withnetwork coding, which prove the correctness of transformation data-oriented service modelpresented in this paper.And then, aim at difficult problem of modeling traffic flow induced by more randomdata flow burst, a new admission control algorithm is proposed, statistical network calculusis adapt to analyze flow performance, admission control based on above analysis results isperformed. Because flow possesses different characteristics in different time scales, andthus have different flow model. We consider two main types of flow models, which areself-similar and multifractal. The presented algorithms select the appropriate time scale byon-line measurement to judge flow burst level and then appoint a flow model. In addition,multifractal flow model can not be directly used for statistical network calculus analysis, soeffective envelope of multifractal flow model is achieved by conversion theorem betweenthe effective bandwidth and effective envelope to solve the previous problem. Becausethe admission control algorithm itself will impact flow performance, admission decisiontime is also crucial to meet the flow performance. In order to guarantee statistical QoS oftraffic flow, low burst period is estimated. And then admission control is performed duringflow burst period.Finally, a dynamic network routing optimization is realized aiming at the dynamicchange of network resources and network traffic flow. Independent and non-independentcross flow are mainly considered. For independent cross flow, the convolution can beconverted to addition operation in the Legendre domain to reduce the computationalcomplexity of the service curve. Legendre transform is performed for arrive curve andservice curve, statistical delay and statistical backlog of traffic flow are expressed by virtueof arrive curve and service curve in Legendre domain, a closed expression of therelationships between them is achieved. This method is applied to the QoS routing optimization algorithm. For non-independent cross flow, due to more than one traffic flowin the same path and overlapping path for different traffic flow, PMOO phenomenon mustbe considered for computing network remainder service curve. Combining with theremaining service curves, relationship between the different traffic flows can be givenby analysis of service curve in Legendre domain, and then corresponding expresson isachieved, which provide foundation for the realization of the dynamic routing considerPMOO phenomenon.