Research On Energy Efficient-based Service Provisioning Techniques In Cloudlet Environment

With the development of compute-intensive application on mobile device,there are big differences between mobile devices and desktop computers(or servers)on processing capacity,storage capacity and battery life.The resource constraint problems have great influence on hardware performance and battery capacity of mobile device when the compute-intensive applications and data-intensive applications are executed.However,mobile cloud computing is an effective method for solving resource constraint problem of mobile device.Compared with other two mobile cloud computing modes,cloudlet has better stability and reliability than micro-cloud,and offers better real time performance than remote cloud.Therefore,the researches of improving cloudlet services,mobile user satisfaction and system performance have important theoretical and practical significance.In this dissertation,energy-effective service provisioning technologies for compute-intensive application in cloudlet which is one special kind of mobile cloud are researched.Aiming at the problems of mobile resource insufficiency,compute-intensive diversity and cloudlet service scope limitation,this dissertation studies energy efficient-based service provisioning techniques in cloudlet from elastic partitioning,opportunistic offloading,seamless handoff and fault-tolerant scheduling.The main research content and contributions of the dissertation are described as follows:(1)The elastic partitioning method for cloudlet application based on graph clustering is presentedSince the existing application partitioning methods based on cloudlet seldom consider mobile device context information and task similarity,it often results in imprecise partitioning.Most application partition granularity is based on category,so it is not suitable for complicated cloudlet environment.Then the graph clustering based cloudlet application elastic partitioning approach is proposed(GCCEP).The proposed partitioning approach uses weighted object relation graph to represent cloudlet application behavior.The tasks are aggregating coded by calculating task similarity from graph clustering feature matrix.Then the elastic partition granularity is decided by task relevance cohesion coefficient and task reuse cohesion coefficient according to mobile device context information.The task coupling is analyzed by coupling relevance matrix and the task balance is controlled by estimating task execution time.The experiment results show that compared with benchmark algorithms,the GCCEP algorithm can reduce partition cost and improve application partition precision and offloading efficiency.(2)The offloading algorithm for cloudlet users based on particle swarm optimization with mutation operator is designedThe existing cloudlet offloading strategies have not considered offloading chance selection and the success probability of offloading,which will lead to problems of frequent offloading,resource wasting and energy efficiency reducing.Thus,the cloudlet user offloading method based on the particle swarm optimization with Mutation Operator is designed(OM-PSOMO)in the dissertation.According to CPU utilization and memory utilization,the proposed offloading method gets overloading opportunity by robust locally weighted regression algorithm.After detecting overloading opportunity,the success probability of offloading is predicted by analyzing the period that mobile users stay within the communication range.Then the tasks are decided whether to be offloaded immediately or be delayed.By calculating response time and mobile device energy consumption,the offloading model is established.Moreover,the optimal offloading algorithm for cloudlet users is designed by using particle swarm optimization with mutation operator.The experiment results show that compared with benchmark algorithms,the OM-PSOMO strategy can reduce offloading cost and energy consumption,and also guarantee mobile user QoS.(3)The fault-tolerant scheduling for cloudlet real-time tasks based on replica constraint is suggestedThe existing fault-tolerant scheduling methods haven’t considered combining virtual resource layer fault-tolerant with task layer fault-tolerant and optimizing replica number,which will increase system cost.The fault-tolerant scheduling for cloudlet real-time tasks based on replica constraint(FTSRL)is suggested.The proposed fault-tolerant scheduling considers above-mentioned problems and models cloudlet virtual machine state by using Hidden Markov Model.The probability of fault state in virtual machine is predicted from virtual machine state transition matrix and probabilistic transfer matrix of output state.The virtual resource layer fault-tolerant is rely on shadow page table.On the other hand,the proposed fault-tolerant scheduling quantizes reliability of cloudlet task.The lower limit of task replica number is calculated according to task fault probability.Considering real time task feature,primary task scheduling and backup task scheduling are proposed.The experiment results show that compared with benchmark algorithms,the FTSRL strategy can improve fault-tolerant ability of the cloudlet system,reduce the redundancy and cost.(4)The seamless handoff scheme for cloudlet service based on handoff gain function is proposedThe existing seamless handoff schemes for cloudlet service consider only horizontal handoff between cloudlets or vertical handoff between cloudlet and remote cloud.This will lead to service interruption,even service failure.Most service handoff methods have problems like frequent handoff and huge energy consumption.The cloudlet service seamless handoff scheme based on handoff gain function(CSHGF)is proposed.The proposed seamless handoff scheme combines horizontal handoff with vertical handoff.The handoff timing is got by analyzing signal strength of mobile device and distance between cloudlet and mobile device.In horizontal handoff,the signal strength of new cloudlet and distance between samples can decide whether to optimize handoff or force handoff.When mobile user moves out of cloudlet service scope,it will execute vertical handoff between cloudlet and remote cloud.In vertical handoff,the optimal network handoff scheme is selected by TOPSIS.The handoff benefit function and handoff cost function can decide the optimal handoff scheme.The experiment results show that compared with benchmark algorithms,the CSHGF scheme can reduce handoff times and service interruption probability,and also improve service quality.