| With the rapid development of cloud computing,the scale of cloud data centers,which are an important part of cloud computing,continues to expand.Data center resource management is confronted with problems of high energy consumption and low utilization,which causes severe carbon dioxide emissions and waste of resources.The data center has heterogeneous characteristics,in particular,workload tasks have diverse resource requirements and QoS requirements.Proper resource management for the data center is not only beneficial to ensure the benefits of cloud service providers and the interests of users,but also helps to protect the ecological environment.Proper resource management for the data center is not only beneficial to ensure the benefits of cloud service providers and the interests of users,but also helps to protect the ecological environment,which will enable the cloud data center to develop in a sustainable direction.On the basis of fully understanding the heterogeneity of data centers,this thesis classifies and analyzes workload tasks,and proposes a high energy-efficient resource allocation strategy(CBRAS)based on task classification in data centers,which can achieve energy-efficient resource management goals,both reduce energy consumption and ensure QoS.The main contributions in this thesis are:1.A reasonable classification of workload tasks.The current study lacks understanding of the heterogeneity of data centers.There is not enough research on task classification methods for several index parameters of a task in a data center.And the classification of tasks mostly focuses on tasks' demand for CPU,memory and other resources,ignoring task QoS requirements.Therefore,this thesis uses K-means algorithm to cluster workload task tracking data of Google cluster production data center.The classification not only considers the resource requirements of the task,but also considers the QoS requirements of the task,including such indicators as priority,delay sensitivity,and duration.2.A resource allocation scheme based on task classification(CBRAS).Currently,resource allocation schemes often overlook the heterogeneous nature of data centers.Therefore,compatibility between task requirements and configuration resources cannot be guaranteed,resulting in higher power consumption and QoS conflicts.Therefore,in this thesis,based on the classification of tasks,according to the differences in the requirements of the tasks,considering the load status of the resources,the resources are configured in a targeted manner.Separating long and short tasks,avoiding the fact that many long tasks are in the process of execution and short tasks have been completed,resulting in increased costs of resource reconfiguration.Configure the same resource queue for tasks that differ in CPU and memory requirements to improve free resource utilization and avoid the phenomenon of resource preemption.At the same time,when the resource is allocated,the task class that guarantees high priority and high time delay sensitivity is priority.In addition,through the flexible configuration of virtual machine resources to reduce the running frequency of CPU and reduce energy consumption.Finally,achieve high energy-efficient resource configuration in the data center,which not only reduces energy consumption,but also ensures QoS.3.Experiments verify the CBRAS strategy.This thesis rewrites the code in the CloudSim cloud simulation platform to implement the CBRAS resource allocation strategy.Through the indicators of energy consumption,the number of physical node enabled,the number of migration times and conflicts times of virtual machines,the performance of the CBRAS strategy was verified both in terms of energy consumption reduction and QoS guarantee.Simulation experimental results find that,compared with the traditional resource allocation strategy based on Round Robin and MBFD algorithm,this strategy has a better performance,especially with the increase in the number of tasks,the advantages are more obvious. |
PDF Full Text Request