Research On Task Deployment In Cloud Data Center

In recent years,with the rapid development of the Internet,the scale of global Internet services and the explosive growth of data information,cloud computing and data centers have become an important part of people’s daily life.With the help of virtualization technology,the cloud data center provides users with flexible and scalable computing,storage and network resources through the Internet.Its service model is that users submit task requests to the data center,and the data center analyzes tasks through its task management and resource scheduling modules.features and their resource requirements,and deploys tasks to appropriate compute nodes based on system resource usage.Task deployment is one of the key technologies that affect the overall performance of cloud data centers.An efficient task deployment strategy is of great significance to improving the resource utilization of cloud data centers and ensuring users’ Quality of service(QoS).This thesis focuses on the optimal deployment of cloud data center tasks from two aspects:independent tasks and workflows:1.For the deployment of independent tasks,the hybrid deployment method of online and offline tasks is studied.For the existing research,the hybrid deployment scheme based on static resource reservation and based on absolute "safe" task combination can guarantee the QoS of online tasks,but it is too conservative and leads to low resource utilization;while the hybrid deployment scheme based on online measurement and real-time adjustment is too risky,which leads to problems such as the inability to guarantee the QoS of online tasks.a joint deployment for hybrid tasks based on dynamic resource allocation is proposed,which can guarantee the QoS of online Tasks.In this way,offline tasks can make full use of the temporarily available resources in off-peak periods brought by the time-varying characteristics of online tasks,so as to improve the task carrying capacity of the data center.Specifically,the proposed algorithm firstly divides the time axis into time slots based on the time-varying characteristics of online tasks and establishes its resource demand statistical model.Secondly,based on the established resource demand model and the QoS violation probability of online tasks,the resource reservation threshold of online tasks is determined to guarantee the QoS requirements of online tasks.Finally,according to the online task resource reservation threshold in each time slot of each computing node,the available resources on each computing node in each time slot of the offline task are determined and the offline task is deployed,so that the offline task can make full use of the temporary available resources brought by the time-varying characteristics of the online task as much as possible to improve the resource utilization.2.For the workflow with data dependency and time sequence relationship,the optimal deployment of workflow under its typical description model,the DAG model is studied.Aiming at the problem that the existing cluster-based workflow task deployment algorithm minimizes the communication overhead at the expense of the parallel execution efficiency of tasks in the workflow,which leads to the low parallel execution efficiency of the tasks.The workflow deployment based on graph segmentation algorithm was proposed to optimize the communication cost and task execution efficiency,and the Workflow completion time was significantly shortened.Specifically,from the perspective of graph theory,the proposed algorithm fully exploits the dependencies and parallelism among tasks in the workflow,and improves the classical graph segmentation algorithm-community partition algorithm,subtly realizing the balance between minimizing the communication cost and maximizing the parallelism in the process of task partitioning.Simulation results show that the WS-GS algorithm not only maintains low communication overhead,but also has advantages in workflow completion time and task parallel execution efficiency.