Research On Software Defined Adaptive Routing Service Customization Mechanism

With the rapid development of Internet technology and the continuous expansion of the network scale,various types of network applications are emerging out,and the user commu-nication requirements for network applications are also becoming more and more diversified and personalized.In addition,users pay more attention to their service experiences.Even for the same network application,the user communication requirements are much different with impact of location factors,mental factors and behavior factors.This requires that the current Internet to be capable in providing diverse and specialized network functions on the commu-nication paths of different network applications,so as to satisfy different requirements with special packets processing and forwarding actions.However,the special network functions usually run on distributed dedicated physical devices in the current network environment.Oriented to the user more and more complicated and frequently changing communication requirements,Internet Service Provider(ISP)has to keep purchasing,deploying and operat-ing new network equipments,which not only causes high capital expense and operating ex-pense,but also brings many difficulties and serious challenges for the service rapid innova-tion,real-time optimization and accurate maintenance.ISP needs a sustainable self-adapting method which can satisfy the user changeable high communication requirements with low investment and good time-to-value.Therefore,based on software defined idea,this disserta-tion proposes utilizing Dynamic Software Product Line(DSPL),Software Defined Network-ing(SDN)and Network Function Virtualization(NFV),in order to enable various instantiat-ed network functions to run on the standard generalized network equipment and to support uniformly,dynamically and flexibly allocating,selecting and deploying diverse network re-sources and functions under the global view.The customized routing services with different characteristics can be composed on the application communication paths by the ser-vice-oriented software engineering method of reusability,assimilability and customizability,so as to satisfy the user special communication requirements and optimize the user service experiences.The main innovations and contributions of this dissertation are summarized as follows:(1)Oriented to the large-scale users' diversified and personalized communication re-quirements,it obviously leads to high capital expenditure and excessive resource consump-tion for the ISP by independently customizing each routing service to each individual user special requirements.This dissertation proposes an adaptive routing service mass-customization mechanism,which can produce large-scale routing services for large-scale users,meanwhile satisfy each user's special communication requirements.Multi-ple Routing Service Product Lines(RSPLs)are established based on DSPL.A feature mode and a consistency orthogonal variability model for RSPL are devised,based on which the large-scale routing services can be rapidly customized,optimized and maintained.The formal definitions and descriptions are formulated to abstract routing service customization,and a method of optimizing ISP profit is devised according to this abstraction.The experimental results show that this mechanism significantly reduces the time overhead of customizing large-scale routing services,and obviously improves the adjustment efficiency and the ad-justment success rate of the running routing services in the real-time.(2)Aiming to address the challenges of service quality optimization,resource allocation optimization and function selection optimization brought to the traditional routing configura-tion pattern by the rapidly increasing in-network big data,this dissertation proposes a big data driven adaptive routing service customization mechanism.A user requirement feature model under multi-dimensional status is established based on DSPL,and it serves as the basis to accurately map and identify the user requirements for the ISP.A dependency model of the user experiences on the requirement parameters is established by utilizing Good-Turing Es-timation,and it serves as the basis to select appropriate network functions and real-timely adjust running routing services.The schemes of the ISP preference evaluation and the user preference evaluation are respectively developed,and the game strategies of service compo-sition and service pricing are designed to achieve the benefits win-win of the ISP and the user according to the schemes.The experimental results show that this mechanism effectively ob-tains the experiences-parameter values distribution and the dependence distribution of the experiences on requirement parameters.In addition,this mechanism significantly improves the ISP and the user satisfaction degrees for the customized routing services.(3)In order to address the challenges of selecting appropriate functions from a large number of candidate network functions to customize routing services,and it also calls for enabling the selection and customization processes to continuously and adaptively optimize,this dissertation proposes a sustainable learning and optimizing enabled adaptive routing ser-vice customization mechanism.A multi-granularity function feature model based on DSPL is established,it serves as the basis of routing service learning and optimizing.A routing service learning scheme is proposed with the machine learning idea introduced.Based on multi-layer feed forward neural network,the routing service off-line learning pattern and on-line learning pattern are respectively devised,and they sustainably optimize network function selection and routing service customization according to the historical information and the user feed-back information,so as to continuously improve the user service experiences.The experi-mental results show that the proposed mechanism significantly improves the utilization rate of diverse functions and the optimization rate of customized routing services with the number of candidate function increasing.In addition,this mechanism rapidly optimizes the user ser-vice experiences.(4)Oriented to the complicated benefit demand relationships among multiple network activity participants(i.e.network operator,ISP,user)under the commercial network opera-tion environment,this dissertation proposes a market driven adaptive routing service custom-ization mechanism.A user utility evaluation model is devised not only to serve as the stand-ard for matching the user preferential services,but also to serve as the reference for the ISP to customize and provide candidate services.A service pricing strategy is devised for the ISP to serve as the basis to optimize the ISP profits by setting reasonable prices.An efficient match-ing algorithm between multiple-applications and multiple-services is devised,and the Pareto Efficiency is introduced in the algorithm to optimize the matching results and improve the benefit equilibrium between the ISP and the user.The experimental results verify the stability and adaptabity of the proposed mechanism,which significantly improves the application ac-cess success rate and the service adaptation rate.In addition,this mechanism also effectively balances the benefits of the ISP and the user.(5)It is impossible to pre-deploy all available network functions into each switch to achieve quickly composing any customized routing service on any communication path at any time due to the switch limited processing capacity.And just immediately deploying most of the real-timely required functions will lead to serious problems such as delay and conges-tion.This dissertation proposes an adaptive deployment mechanism of network functions in customized routing services.A network function pre-deployment scheme based on prediction is proposed by combining long-term perdition and short-term prediction to pre-deploy the functions,which are predicted to be largely required in the future,into each switch.A net-work function real-time deployment scheme based on network status is proposed with explic-it consideration of three factors including the switch limited processing capacity,the an-ti-affinity among functions in a switch and the link real-time available bandwidth to improve the resource utilization rate.The experimental results show that this mechanism is good at predicting the function popularities in the future,and significantly improves the required function hit rate in each switch.It also reduces the time overhead of composing customized routing services.