Heterogeneous Resources Allocation In Mobile Edge Computing

In recent years,the rapid development of mobile Internet and Internet of Things has given birth to many new applications.Among these,computation-intensive applications,such as video transcoding,3D online games,and reality augmentation,generally consume a large amount of computation resource,result in large energy consmption,and require very short response delay.However,limited by computation,storage,and energy resources,terminals often could not handle computation-intensive applications independently.Traditional method is to offload computation-intensive applications to cloud computing center to be processed.However,this method not only brings out a large number of data transmission which puts great pressure on bandwidth of uplink and downlink,but also results in long latency and large communication energy consumption which deteriorates user experiences.In this context,Mobile Edge Computing(MEC)comes into being.By equipping edge servers with computation and storage resources,MEC could provide computing environment with neardomain characteristic,which could overcome shortcomings of cloud computing,e.g.,large delay,huge energy consumption,and bandwidth pressure.Different from simple communication and computation procedures,computation offloading involves both communication and computation resources,and brings out a tradeoff between communication cost and computation cost.Any unreasonable offloading strategy degrades the gain in energy consumption and delay,even results in performance loss.Hence,how to design resource allocation to well tradeoff communication and computation cost is the key to achieve high gain of MEC systems.Additionally,different from traditional cloud computing center,edge servers are highly limited by communication,computation,and storage resources.When a single edge server could not handle computation-intensive applications efficiently,multiple edge servers should perform collaborative computing.Take video service in the 5th Generation mobile communication(5G)as an example.Video transcoding essentially changes video delivery mode from “cache and transmit”into “cache/process and transmit”,and collaborative computing further enriches diversity of video retrieving mode,which improves efficiency of video caching.However,there is no fixed strategy for how to collaborate,and different collaboration strategies lead to different gains and costs.Hence,it is a key challenge to design efficient collaboration strategy under heterogeneous video requests in each edge server,to balance communication,computation,and storage loads among multiple edge servers.As motivated,this dissertation investigates heterogeneous resource allocation in MEC system from two aspects of computation offloading and collaborative computing.Here,heterogeneous resources refer to communication resource,computation resource,and storage resource.In computation offloading scenario,this dissertation investigates a unified allocation method of computational speed and communication power,and a dynamic matching method of individual offloading and group resource allocation to improve user experiences,such as reducing energy consumption and delay.In collaborative computing scenario,this dissertation investigates a unified adjusting method of video caching and delivering to minimize content access delay.Main achievements of this dissertation are summarized as follows:1.Aiming at single user computation offloading systems,a unified allocation method of computational speed and communication power is proposed to minimize terminal energy consumption and improve user delay experience.Specifically,parallel computing between terminal and edge server would result in coupling feature of offloading operation.By mining this feature,and modeling computation and communication overhead respectively,this dissertation formulates terminal energy consumption minimization(ECM)problem and latency minimization(LM)problem as non-convex problems.In order to solve these two problems,two algorithms are designed based on the method of variable substitution and univariate search technique.Different from traditional algorithms,the algorithms designed in this dissertation could fully use Dynamic Voltage Scaling(DVS)technology to reduce terminal energy consumption and latency of application execution.In addition,we theoretically prove that algorithm designed for ECM problem could reach the global optimal solution,and algorithm proposed for LM problem is local optimal.Finally,simulation results verify correctness of theoretical analysis and effectiveness of proposed algorithms.Note that this work is a fundamental research in computation offloading,which reveals conditions to achieve gain of offloading,and provides theoretical guidance for the study of computation offloading in more complex scenarios.2.Aiming at multiuser computation offloading systems,a dynamic matching method of individual offloading and group resource allocation is proposed to minimize weighted sum of mobile energy consumption.Specifically,besides being coupled due to parallel computing,offloading operations in multiuser computation offloading systems are further coupled due to competing for limited resources.By digging deeply into this dual coupling feature,we formulate the weighted sum of mobile energy consumption minimization problem,and design a high-performance and low-complexity offloading algorithm.By iteratively optimizing offloading ratios of each user and allocating resources among multiple users,the proposed algorithm could finally achieves good matching between individual offloading behavior and multiuser resource allocation,thus reducing weighted sum of energy consumption of users in the system.Finally,simulation results verify effectiveness and low complexity of proposed algorithm,which could be well applied to actual systems.Compared with other algorithms,proposed algorithm could obtain better matching between individual behavior and group competition,and thus reduce energy consumption of terminals.3.Aiming at a scenario where multiple edge servers collaborate to provide video services,a unified adjusting method between video caching and delivering is proposed to minimize content access delay.Specifically,by digging into coupling feature between video caching and delivering,we jointly consider caching strategy on slow time scale span and video delivering strategy on fast time scale span,and formulate content access delay minimization problem as a two-time scale stochastic integer linear programming problem.Furthermore,a two-step algorithm based on sample average approximation is proposed.This algorithm first designs video caching strategy based on arrival statistics of requests and expected video delivering strategy,and then designs video delivering strategy based on real requests for each fast time slot.Finally,simulation results verify advantages of proposed algorithm in reducing content access delay and improving cache hit ratio.