Task Offloading And Resource Allocation In Vehicular Edge Computing

Intelligent Transportation System(ITS)applies a variety of sensors,communications,image recognition and other technologies to traffic management,and creates convenient and safe intelligent transportation services.The Intelligent Connected Vehicles(ICV)also enable vehicular users to make high-speed and reliable data connections,and promote communication from vehicles to road traffic infrastructure or other equipments.This makes intelligent applications such as autonomous driving,intelligent navigation,route planning and in-vehicle entertainment services possible.However,existing resource-constrained vehicles cannot fully meet the low / ultra-low latency requirements of these applications.By offloading some of the vehicles’ compute-intensive tasks to edge servers,Vehicular Edge Computing Networks(VECN)are regarded as a promising technology.This paper studies the applications and challenges of mobile edge computing in assisting vehicles to perform intensive computing tasks,and proposes effective solutions.First of all,by using the characteristics of high capacity,low latency and high reliability of optical fiber networks,ubiquitous connectivity and multiple wireless access technologies support are provided for vehicular users.After that,this paper proposes a Software Defined Networking(SDN)based Fiber-Wireless(FiWi)enhanced VECN,in which vehicles’ Synchronous Random Walk Model(SRWM)is considered,in order to accurately simulate the real-world road traffic.In the above scenario,this paper studies how to reduce the processing delay of vehicles’ computation tasks by computing offloading.We propose Enumeration-based Optimal Task Offloading Algorithm(EOTOA)that can obtain the optimal solution to the problem,however,it has extremely high computational complexity.In order to solve this problem,Game Theory based Nearest task Offloading Algorithm(GTNOA),Predictive Game Theory based task offloading Algorithm(PGTOA)and Approximate Load Balancing task Offloading Algorithm(ALBOA)are proposed.The experimental results show that both GTNOA and PGTOA can obtain the approximate optimal solution of the problem,but they can not guarantee the load balancing of computing resources between the edge servers,and ALBOA can obtain the approximate optimal solution of the problem while ensuring the load balancing.In order to cope with the highly dynamic network communication and computing resources environment in VECN,and provide flexible task offloading solutions.this paper further studies the application of reinforcement learning in task offloading,and proposes Q-Learning based task Offloading Algorithm(QLOA)and Deep Q-Learning based Task Offloading Algorithm(DQLOA).Both QLOA and DQLOA can obtain approximate optimal solution of the problem,and have better flexibility than ALBOA.In addition,DQLOA has higher execution efficiency than QLOA.Finally,in order to further reduce the tasks’ processing delay,this paper studies partial offloading of vehicles’ tasks.We first use the Unoffloadable Subtask Merging Algorithm(USMA)to merge subtasks that cannot be offloaded to the edge servers,to simplify the complexity of the partial offloading problem.After that,Enumeration-based task Partial Offloading Algorithm(EPOA)and Deep Q-Learning based task Partial Offloading Algorithm(DQLPOA)are proposed to obtain optimal partition configuration.The final experimental results prove the good performance of DQLPOA and the importance of partial offloading for task processing delay reduction.