Research On Bi-objective Optimization Models And Algorithms For Service Function Chainings Mapping In Elastic Optical Networks

Due to increasing diversity and incompatibility of traditional network hardware devices,it is increasingly hard for network service providers and data center operators to provide technical support for updating and maintaining these devices conveniently.Correspondingly,network function virtualization(NFV)technology appears.It decomposes many complex network devices into a variety of virtual resources with virtualization technology.These virtual resources can be flexibly deployed on data center server,which effectively solves the increasingly difficult problem in updating and maintaining network functions.The mapping problem of service function chaining(SFC)is a basic problem when realizing NFV technology.Its main task is to obtain the best resource allocation scheme by optimizing the criterions of SFC mapping.In addition,due to continuous development of large data and cloud computing technologies,traffic gradually presents the characteristic of “elephants flow” and “mice flow”,which will result in network loads unbalanced and more requests blocked.However,elastic optical networks(EON)can flexibly allocate resources for users according to their traffic granularity,which greatly reduces the blocking ratio of SFC requests.This thesis focuses on the mapping problem of SFC in EON,and its main work and contributions are as follows:(1)The problem about the network resource overhead of SFC mapping is investigated: First,a mathematical model of SFC mapping in data center over elastic optical network(DC-EON)with unlimited resources is established.This model aims at minimizing the total cost of normalized network resources overhead.Second,to realize the deployment of virtual network functions(VNFs)and the routing and spectrum assignment(RSA)of virtual links,node betweenness centrality and link betweenness centrality based on traffic-awareness are defined respectively.The deployment of VNFs is based on traffic related node betweenness centrality.The RSA scheme of each virtual link trades off three parameters,the requested bandwidth,the length of candidate path,and the link betweenness centrality of links on candidate path.Then,a betweenness centrality based comprehensive SFC mapping(BCCSM)algorithm is proposed to reduce resource usage effectively.Compared with the existing longest common subsequence based algorithm(LBA)and the shortest-path batch VNF deployment algorithm(SBA),BCCSM obtains the least network resource overhead.(2)The bi-objective optimization problem of SFC mapping in finite resources DC-EON is studied.First,the constraints of server resources and bandwidth resources are add into the process of mapping SFC.With the optimization objectives of minimizing both the number of VNFs deployment and maximum frequency slot number occupied on all links,a mathematical model of SFC mapping is established.Second,to solve the proposed model,a feedback adjusted differential evolution operator is designed to adjust the scaling factor adaptively.Then,a differential evolution algorithm with feedback adjustment based on decomposition(DEFAD)is proposed,which solves the resources allocation problem of SFC mapping in DC-EON with limited resources from the perspective of multi-objective optimization in the first time with my best knowledge.The simulation results show that compared with the existing algorithms,DEFAD could get the resource allocation scheme closer to the actual pareto frontier,and obtain multiple optimal solutions under of different objective preferences at the same time.In view of service requirements like transmission capacity and reach,the practical EON has the characteristics of layered and multi-domain structure,so the resource allocation problem of SFCs mapping needs further study.In addition,when dealing with the SFCs mapping problem,the proposed DEFAD algorithm uses uniform weight vector initialization,which causes the uneven and poor diversity problem of pareto solutions due to the non-equivalence of two objectives.The next step is to combine the uniform design,orthogonal design and other initialization methods with intelligent algorithms to solve the problem.