Research On Incentive-driven Data Offloading Through Opportunistic Mobile Networks

With the rising popularity of smart devices and wireless services,the mobile traffic in mobile network is growing explosively.Therefore,there is an imminent requirement for the Content Service Provider(CSP)to provide quick and efficient ways to ease the traffic load of cellular networks.Mobile data offloading is regarded as an effective way to ease the burden of traffic in cellular networks,which applies complementary network communication technologies to deliver mobile traffic that was originally planned to be transmitted via cellular networks,e.g.,Wi-Fi networks and Small Base Stations.However,these networks all depend on infrastructures,which has the disadvantages of limited coverage and high installation that limit their use and promotion.Another effective way is to offload cellular traffic via Opportunistic Mobile Networks(OMNs),also called opportunistic offloading,which uses the mobility and caching capacity of nodes to assist in data offloading,so as to reduce cellular traffic and thus the operation cost of the CSP.However,users may not be willing to provide data offloading services without receiving proper financial incentives since it will cause additional resource consumptions inevitably,e.g.,energy consumption,capacity consumption,etc.Therefore,it is important to exploit effective incentive mechanisms to stimulate users in OMNs to provide data offloading services.Although there are many researches on data offloading via opportunistic mobile networks,and some of them have considered the corresponding incentive mechanisms,they are far from enough.Compared with the existing work,this paper has established a more complex and realistic system model,and formulates the reverse auction from the perspective of the CSP,which can ensure the individual rationality,truthfulness and optimize the benefits of the CSP from the business point of view.This dissertation studies the incentive-driven data offloading through OMNs.The main research work includes:(1)The single content opportunistic offloading problem based on reverse auction is studied.The incentive-driven data offloading through OMNs is modeled as a Non-Linear Integer Programming problem from the business point of view,aiming to minimize the cost of the CSP.Furthermore,to improve the performance of data offloading,a Decay-based Helper Selection Method is proposed to select far apart nodes with higher offloading potential and less payment as helpers.Finally,a large number of real trace-driven simulation results demonstrate the superiority of the proposed decay-based helper selection method under different scenarios.(2)The opportunistic offloading and caching management based on reverse auction is studied.Based on the previous research,the single content model is extended to the multicontents,and the offloading process is divided into multiple time slots.The incentive-driven data offloading process is modeled as a Non-Linear Integer Programming problem from the business point of view,then a Greedy-based Helper Selection Method and a Caching Replacement Scheme are proposed to solve the optimization problem.In addition,an innovative payment rule based on Vickrey-Clarke-groves(VCG)model is proposed to ensure the individual rationality and authenticity of the proposed algorithm.The real trace-driven simulation results prove the superiority of the proposed Greedy-based Helper Selection Method under different scenarios.(3)Multi-contents opportunistic offloading and caching management based on reverse auction and deep reinforcement learning is studied.Based on the previous research,the contact of nodes is modeled as Pareto distribution,then the data offloading and content caching process is modeled as a Integer Non-Linear Programming problem from the business point of view.At the same time,a deep reinforcement learning technology is introduced and a content caching method based on Deep Reinforcement Learning is proposed to get the approximate optimal solution.Extensive real trace-driven simulation results demonstrate the superiority of the proposed algorithm.