Low Delay Computation Offloading In Mobile Edge Computing Systems

With the advent of 5G era and the prosperity of Internet of Things(Io T),many emerging computation-intensive mobile applications,e.g.,augmented reality/virtual reality(AR/VR)and in-telligent manufacturing,are realized on intelligent mobile terminals(MTs).These applications pose exigent requirements on the quality of computation experience,which cannot be easily satis-fied by solely relying on MTs.Computaion offloading in mobile edge computing(MEC)systems,which offloads the computing-intensive computation tasks to the MEC servers deployed in close proximity,is emerged as a promising technology to handle the explosive computation demands and the everincreasing computation quality requirements.Researchers from both academia and industry have been actively promoting MEC technology by pursuing the fusion of techniques and theories from both disciplines of mobile computing and wireless communications.However,com-pared with the increasing computing demand,the limitation of MEC servers'computing resources is becoming obvious.Also,with the expansion of the network scale,the distribution of computing resources and demands will be unbalanced.Therefore,how to design and optimize the computation offloading approach to guarantee the low delay of computation tasks is an important technology.Motivated by this,this thesis presents some researches on the low delay computation offloading in MEC systems.Firstly,for small-scale MEC systems,this thesis considers the low delay computation offload-ing in single-hop MEC systems.We develop an analytical framework for delay-optimal compu-tation offloading in computation-constrained MEC systems,and derive a closed-form offloading policy.Specifically,We consider the computation-constrained MEC server for the delay-optimal computation offloading problem.In this system,the computation delay of the MEC server can-not be neglected,and the cascade queue balance should be maintained.For achieving good delay performance,the delay-optimal computation offloading policy should jointly consider the channel state information(CSI),the local queue state information(LQSI)and the remote queue state infor-mation(RQSI)simultaneously.We formulate the delay-optimal computation offloading problem as an infinite horizon average cost MDP,and adopt a virtual continuous time system(VCTS)with reflections to overcome the curse of dimensionality.Next,we develop a multi-level water-filling computation offloading policy for jointly considering the CSI,the LQSI and the RQSI.Then,we derive the dynamic instantaneous rate estimation for maintaining the cascade queue balance by estimating the in-out rate difference of the queue system.Finally,we obtain approximate priority functions in both the computation sufficient scenario and the computation constrained scenario.We extend our policy to the multi-MT multi-server scenario by adopting learning approach.Specifical-ly,we compare the main differences between two scenarios,and derive a computation offloading policy by learning the access ratios from the historical access records.Secondly,for large-scale MEC systems,this thesis considers the low delay computiation of-floading in multi-hop MEC systems.We first investigate the back-pressure-based dual-properties queue control method in distributed manner,and achieve the optimal queue control method un-der different resource constraints.Then,we investigate the computation offloading with a novel framework for achieving a balanced distribution of computing resources and computing loads,and provide the Communication-Computation Throughput Optimal Task Offloading(CTO~2)ap-proach.Specifically,we first formulate the optimization problem by adopting Lyapunov optimiza-tion.For achieving communication-computation throughput optimal,we jointly consider the task scheduling,power allocation,and link scheduling problem in the MEC system.We decompose the master problem into three parts with a nested structure,analysis the relationship between three sub-problems,and solve each sub-problem step by step.We obtian the optimal task scheduling by solving an combinational problem,derive the optimal power allocation by adopting Dual Con-sensus Alternating Direction Method of Multipliers(DC-ADMM),and achieve the optimal link scheduling through L-CSMA approach.Finally,following the nested relationship between three sub-problems,we achieve the CTO~2approach in distributed manner.In the end of this thesis,we summarize the mainly contribution of this research,and prospect the future work of low delay computation offloading in MEC systems.