Researches On Power Collection And Virtual Environment Power Evaluation Of Data Centers

In recent years,the number and scale of data centers are increasing,leading to more power consumed by data centers.This arouses widespread concern from all walks of life.The power consumption of IT equipment is the main part of data center power consumption,and the power consumption of IT equipment is mainly produced by physical machines.The virtual environment running in the physical machines is the main cause of physical machine power consumption.Virtual resource scheduling is a crucial task in data centers,and the power consumption of virtual resources is a critical foundation of virtualization scheduling.The virtual environment includes VMs(Virtual Machines)and containers,but the power consumption of them cannot be measured.Therefore,it is imperative to establish a multi-level power estimation algorithm of VMs and containers.This paper starts from the aspects of VM and container and discusses the multi-level power estimation problem.From the physical machine level to the VM level: Due to the dynamic change of the number of VMs in the physical machine,this paper adds the feature vectors of the same type of VMs together to fix the number of features,avoiding the huge overhead of retraining the model.At the same time,in order to effectively model the nonlinear relationship between features and power,the paper proposes a piecewise linear model based on decision tree and tests it under different tasks.The experimental results show that the algorithm achieves better performance than existing algorithms under different tasks and time granularity.From the VM level to the container level: If the container power estimation model is established on the basis of the VM power,it may bring greater errors due to twice modeling.According to the problem,this paper regards the container as a group of processes running in the host machine and substitutes the container features that correspond to the VM features into the VM power model to estimate the power of the container in the VM.Relevant experimental results show the validity and effectiveness of the algorithm.Due to the lack of public data center power consumption datasets,this paper also gives a detailed introduction to the establishment of the simulation environment,data collection,and data storage.In addition,an algorithm library for data center power estimation and prediction is developed.The system module integrates the algorithms proposed in this paper and some popular algorithms,which can provide a timely and accurate reference for data center management.