Research On Dynamic Resource Management Based Vehicular Cloud Networks

In recent years,an emerging technology called Vehicular Cloud Networks(VCN)has developed with the advance of Internet of Vehicles and Cloud Computing.In VCN,resources(e.g.,storage,computation)of vehicles and infrastructures are integrated into the virtual resource pool.Cloud system can dynamically allocate storage resources and computing resources for user service based on the available computing resources in resource pool and the changes of the request.Simultaneously,this is in effort to fulfill maximum utilization ratio of the resources and the improvement of processing speed.The mobility of the vehicles caused the complicated dynamic change of total amount of virtual resources in VCN resource pool,and this new problem bring serious challenge to the existing cloud computing resource allocation.Under VCN with the two-layer cloud architecture(Local Cloud and Central Cloud),different layer of cloud has its own characteristics.The local cloud is composed of vehicles and roadside infrastructures,so its computing resources are relatively limited and changing with the number of resource vehicles.However,the local cloud has low transmission delay to the vehicle users;the central cloud is composed of remote servers and other infrastructures,which has rich computing resources but experience high end-to-end delay.This paper first studied the cross-layer resource management based vehicular cloud networks.Then,under VCN with two-layer architecture,we used Semi-Markov Decision Process(SMDP)to optimize the cloud computing resources management.This thesis proposes the CL-SMDP(Cross Layer-Semi-Markov Decision Process)resource management scheme to obtain the maximal long-term reward for providing cloud service.Finally,by comparing with the traditional VCN resource management based on Greedy Algorithm(GA),the simulations indicate that the CL-SMDP resource management scheme can not only significantly improve the long-term reward of the system,but also effectively reduce the rejection rate of the requests.In addition,this thesis also studies the cross-area resource management in VCN.Considered the imbalance of loads and resources in different cloud service domains which are caused by the movement of vehicles,we propose the CA-SMDP(Cross Area-Semi-Markov Decision Process)resource management scheme.Due to considering the service migration,local cloud resource management should additionally consider the request of the local users and the requests of the neighborhood cloud users.The impact of different types of service requests on system reward is also different,so we introduce SMDP for finding the reasonable strategy of resource allocation,and obtaining the maximal long-term reward.Simulation results indicated that the proposed algorithm could highly improve rejection rate,migration rate and the long-term reward of system by comparing with GA and CL-SMDP.