| Cloud environment provides many on-demand high-performance computing services,and more and more data-intensive workflows are deployed to run in cloud computing environment.Due to the different functional requirements of each task in the workflow,these interrelated tasks may be assigned to different cloud service providers for execution.The massive amount of data to be transferred among tasks greatly increases the execution time and cost,and reducing the total amount of workflow execution transmission has become the core index of many scheduling methods.However,the amount of data transmission between tasks in some workflows cannot be accurately estimated before execution,which makes it difficult for traditional static scheduling methods to obtain effective scheduling schemes based on uncertain transmission volume information.In order to solve the scheduling problem of the above uncertain data-intensive workflow in cloud environment,this thesis proposes a dynamic task allocation method based on data dependency.The main work includes:(1)Pretreatment: workflows with typical logic such as sequence,parallelism and selection are modeled in the form of process algebra,and a set of mutually exclusive tasks is obtained through binary tree traversal.Then the maximum data path set of each task is obtained according to the data dependencies of each task in the workflow.(2)Based on the mutually exclusive task set of the tasks to be allocated and the sets of maximum data paths,considering the amount of generated real-time data to be transferred from the executed tasks to a certain task to be allocated and the information of all candidate service providers of the tasks related to the certain task to be allocated,a reasonable candidate service provider is dynamically selected for the task to be allocated to reduce the total amount of the transferred data of the entire workflow.In this thesis,the feasibility of the scheduling method is verified through a specific example,and a comparative experiment is designed and implemented to demonstrate the method proposed in this thesis can generate a better solution with less total amount of workflow execution.(3)A workflow scheduling system is constructed according to the above preprocessing method and dynamic allocation method.The system mainly includes a random workflow generation module and a comparsion scheduling module.The workflow in the experiment is directly and randomly generated by the randomly generated workflow module.The comparsion scheduling module implements three different comparsion methods and the method in this thesis.The scheduling results show that the method in this thesis has a better saving effect on data transmission.This thesis proposes a dynamic task allocation method for data-intensive workflow in cloud environment where it is difficult to predict the amount of transmitted data in advance,and demonstrates its feasibility through an example.The effectiveness is verified through simulation experiments,which can reduce the total amount of execution and transmission of this type of cloud workflow and indirectly optimize the execution cost and completion time. |
