Optimization Of Computation Offloading Based On NFV In Fog Computing Networks

The number of mobile smart devices has increased significantly,but the CPU processing and battery capacity of those devices is limited to support computation insentive services.Fog computing is a feasible solution to improve the quality of user experience and alleviate backhaul bandwidth consumption caused by mobile cloud computing,by expanding cloud case and deploying computing capabilities at network edge nodes.Network Function Virtualization provides more efficient,flexible and scalable network function services for fog computing,since it moves network functions from dedicated hardware devices to general-purpose servers.Therefore,fog computing based on NFV reduces APEX and COPEX of oprators by flexibly utilizing network and computing resources.This thesis focuses on the optimization of computation offloading based on NFV in fog computing networks.The research work mainly includes:Firstly,optimization of computation offloading in fog computing based radio access netwok is investigated for online game applications.Considering the uncertainty of user behavior,a probabilistic service function chain model is proposed which are provided physical resources by fog computing wireless access network.The optimization problem is modeled as an integer programming problem with the objective of minimizing the system cost.In order to reduce the algorithm complexity,a heuristic algorithm called PSECO is proposed and evaluated.Secondly,taking into consideration on user mobility and characteristic of face recognition application accessing the database,a multi-objective Collaboration optimization algorithm of Edge caching and Computation offloading for Face recognition applications(CECF)is proposed,and the CECF problem can be divided into two sub-problems:edge cache placement optimization with minimal cost(CPMC),and computation offloading optimization with minimal delay,where CPMC is dynamic long-term optimization problem.Based on the Deep Q Network(DQN)algorithm,a Centralized DQN(CQ)algorithm is proposed for the dynamic CPMC,and in order to solve the scalability problem,three distributed algorithms are proposed,namely Local state based distributed DQN(LSDQ),Observable state based distributed DQN(OSDQ)and DQN based Distributed Value Functions(QDVF),respectively.Simulation results show that the proposed distributed algorithms have good performance in terms of cost,hit ratio and delay.