Research On Computational Resource Optimization Of Vehicular Clouds In Multi-tier Cloud-enabled Internet Of Vehicles

With the rapid development of the Internet of Things(IoT)and mobile wireless communication technology,as well as the improvement of people’s living standards and the mass popularization of vehicular terminals,the Internet of Vehicles(IoV)becomes popular as the "nerve center" of the Intelligent Transportation System where the data production speed has far exceeded Moore’s law.Due to the rapid growth of big data and new application modes which are computation-intensive in IoV,the resources and capability of the on-board equipment in a single vehicle terminal can no longer satisfy the powerful computing,analysis and storage capabilities of mass data required by such applications.How to solve the above mentioned problem becomes a hot topic in recent years.This thesis mainly establishes a cloud-enabled IoV system that supports efficient resource management,where the computation offloading and resource optimization issues are studied under the multi-tier cloud-enabled IoV architecture;studies the single as well as multi service requestor based single-site computation offloading and resource optimization、unstructured as well as graph-based multi-site computation offloading and resource optimization,in order to effectively expand the dimensions of applications,enhance the computational capabilities while improving service efficiency and user experience.This thesis proposes a single-site dynamic computation offloading and pricing strategy based on Stackelberg game,where the joint optimization of service provider selection,computation offloading rate as well as pricing is proposed under constraints.In mobile offloading complete information game,the problem is modeled as a common resource allocation problem where the optimal service provider and the corresponding offloading rate can be obtained.In mobile offloading incomplete information game,the Stackelberg equilibrim and the corresponding existing conditions are detailly analyzed.Performance evaluation based on Monte Carlo simulation shows that the proposed methods can reach 78.5%and 74.3%improvement on application completion time when compared to local computing scheme;comparing with first come first serve and the fastest processing schemes,the proposed methods can significantly achieve mutually satisfactory computation offloading decisions.Then,this thesis studies the single-site computation offloading problem of multi-service requestors by enlarging the scope of vehicular clouds,and proposes a Vickrey-Clarke-Groves reverse auction based computation offloading and resource optimization modeled by integer linear programming problem(ILP)with constraints.The optimal solutions can be obtained by undergoing high computational complexity.In order to support larger and fast changing network topologies,an efficient unilateral-matching-based method is proposed,which offers satisfactory sub-optimal solutions with polynomial computational complexity,truthfulness and individual rationality properties as well as matching stability.Performance evaluation shows that the proposed method can improve the system efficiency in various traffic conditions,on application completion time,the value of objective function as well as vehicle-pairs under service.For example,compared with the local computing scheme in low-traffic scenarios(n=m=6),the proposed method can reduce 77.5%of the application completion time while only 5.9%larger than the optimal solution;also,the proposed method can serve more vehicles than fast processing scheme,lowest cost scheme as well as first come first serve scheme.For the multi-site computation offloading and resource optimization problem where a computation-intensive application can be divided into several parts and offloaded to various nearby vehicles under opportunistic contacting between moving vehicles.This thesis proposes a model combining of ∞-norm、1-norm and F-norm under constraints,which is a joint optimization of application completion time,costs of service requestors as well as incentivation of service providers.For solving the Np-hard problem in an efficient manner,the thesis proposes an improved simulated annealing algorithm which can effectively jump out of the local optimum to explore the global optimum solutions.The performance evaluation shows that the proposed method can achieve good performance on application completion time、pricing of service and convergence characteristics.For example,the proposed method can reduce 93.5%and 78.5%(n=3)on application completion time when compared to local computing and average computing schemes;comparing with the average offloading scheme,the proposed method can achieve 12.8%on profit increasing.Finally,this thesis proposes a multi-site offloading method over vehicular cloud by modeling the computation-intensive applications as weighted undirected graphs with respect to opportunistic contacting between vehicles and limited available resources.The offloading problem is formulated as non-linear integer programming problem,aiming to minimize job completion time and data exchange cost while achieving the computation load balancing.In low-traffic scenarios,we determine the optimal solutions by combining traverse with subgraph isomorphism algorithms;in rush-hour scenarios,given intractable computations for deriving feasible allocations,we propose a novel hierarchical tree based randomized subgraph isomorphism algorithm with low-complexity.Numerical analysis and comparative evaluations are performed under different application topologies and vehicular cloud configurations,which indicate that the proposed algorithm can achieve near optimal solutions on both the value of the objective function as well as the successful rate on offloading.For the sake of improving the system performance,the computation offloading and resource optimization methods in dynamic topology can be further studied based on the characteristics of application structures.