Research On The Key Technologies Of Tasks Optimization Scheduling For Cloud Computing

In recent years,as that cloud computing continues to expand high-speed application development and user groups,cloud computing systems need to schedule and manage the users' task.Therefore,how to carry out reasonable and efficient management methods on the various application tasks and maintain the load of the system in a relatively balanced state has become a key issue affecting the performance of cloud computing.At present,cloud computing task scheduling strategies mostly use traditional distributed computing task scheduling methods,but the two distinguishing features make this reference imperfect.For cloud computing technology is in widespread commercial background,the scheduling strategies not only meet the needs of users,but also take into account the interests of the service providers.That feature is not available before the technology in cloud computing.Therefore,researching the task scheduling optimization strategy of cloud computing has important theoretical and practical significance.Based on the characteristics and demand analysis of cloud computing system,this thesis deeply studies the task optimization strategy of cloud computing.Firstly,the architecture of task scheduling in cloud computing environment is given,which provides a global management guarantee for task scheduling.Then,for base layer in cloud computing datacenter,the problem that the task scheduling is inefficient and energy consumption is great needs to be solved.Combined with the application layer in cloud data center,Cloud computing users are look forward for task allocation fairness and cloud service providers are look forward for the effectiveness of mission assignment needs,we respectively propose for cloud computing center base layer and application layer task scheduling models and algorithms.On this basis,considering the efficiency requirements of cloud workflow task scheduling in multiple data centers,a task scheduling optimization strategy with associated dependencies is proposed.The main research work of the thesis is summarized as follows:In order to provide a good cloud computing environment,appropriate fault tolerance measures need to be taken to build a new task optimization scheduling architecture.Different from the design of other cloud computing architectures,the user needs to be given the conditional assumptions.Meanwhile,the new task scheduling framework only needs to instantiate the parameters according to the specific scheduling requirements.At the same time,a dynamic data replication mechanism is introduced into the architecture and DRA(Dynamic Replica Algorithm,DRA)is proposed.With the increase of the number of copies and the load factor,the proposed DRA algorithm reduces system latency and bandwidth consumption comparing with the FRN(Fix Replica Number,FRN)and LD(Local Dynamic,LD)with a fixed number of copies,and as the number of user tasks increases.The DRA algorithm reduces resource consumption for creating replicas based on ensuring shorter average task completion time.Aiming at optimizing scheduling of a single data center in the basic layer in a cloud computing system,users'requirements for cloud computing resources diversity and multi-index constraints is considered.Based on Markov theory,a time-cost model is established,and the task scheduling algorithm of genetic ant colony algorithm is proposed with the new cloud computing task scheduling system architecture based on dynamic replica mechanism.For redesigning the crossover factor in genetic algorithm,the convergence effect of the algorithm in global search is improved.The global search result of improved genetic algorithm is used as the next ant colony optimization algorithm to solve the initial value of pheromone in task optimization scheduling.Finally,Cloud computing data center basic level task optimization scheduling strategy is working.The experimental results show that the task execution time and task execution costs reduces,also that CPU,disk and memory utilization increases when the proposed MGAA(Modified Genetic Ant Algorithm,MGAA)task scheduling strategy is compared with algorithm LGA(Load Balancing Genetic Algorithm,LG A),and BIGA(Buffer Setting Improved Genetic Algorithm,BIGA).In the application layer of a single data center of cloud computing system,due to the market mechanism of cloud services,aiming at the urgent need of cloud service providers to reduce the cost of cloud services and increase the revenue of cloud services,an optimal solution named MRA(Modified Resource Algorithm,MRA)for task scheduling based on Lagrange multiplier method is proposed.Nash equilibrium theory and queuing theory are taken as the theoretical basis of the study.Task queue scheduling model of service response time is established to solve the task scheduling problem.At the same time,in the application layer of a single data center,taking into account the interests of users,in order to ensure resource allocation more reasonable and fair,an adaptive time-aware task assignment model is established and a scheduling algorithm based on gradient projection is proposed The algorithm GP(Gradient Projection,GP)considers the user's execution order and execution time of different application tasks,and uses Nash equilibrium theory to solve the task scheduling problem with priority and time limit.Experimental results show that the MRA algorithm can converge to an equilibrium solution to optimize task scheduling and resource allocation.The proposed GP algorithm has better convergence than FS(Fair Schedule,FS)in the face of different priority tasks,RS(Random Schedule,RS)and EDF(Earliest Deadline First,EDF).The resource utilization of the system is improved,and the execution delay of the task is significantly reduced of GP algorithm simultaneously.In order to solve the problem of unified planning and coordination management of multi-data center tasks with correlated relationship,the task scheduling of multi-data center is divided into two stages.In the first stage,according to the relationship of tasks,combined with the workflow technology,using the DGA(Directed Acyclic Graph,DAG)to represent the association of tasks,an improved MPSO(Particle Swarm Optimization,MPSO)is given.In the MPSO,the particle velocity displacement and weight parameters are improved,and the particle is found to be in the best position to mine the global optimum to be transformed into a multi-objective optimization solution that satisfies the constraints.At the same time,Pareto multi-objective optimization method is introduced to solve the optimization problem with constraints.In the second stage,considering the load situation among multiple data center systems,a task scheduling algorithm with load-aware is proposed.Load-sensing parameters are introduced into the algorithm to adjust the system load during task execution.The experimental results show that,with the task execution in the system,the time and cost spent by the MPSO algorithm using DAG and workflow scheduling in task execution are less than HEFT(Earliest Finish Time,HEFT),SAA(Simulate Anneal Algorithm,SAA).For which the MPSO is partially optimized,and the load-sensing algorithm can raise the resource availability comparing with the system without load-sensing,ensures the efficient execution of tasks in a sequential manner.