Energy Efficiency Aware Virtual Machine Scheduling

The power and energy consumption in today's data centers are becoming the major concern of data center operating and management.In large Internet companies,such as Google,Facebook,Microsoft,and Alibaba,their servers' PUE(Power usage effectiveness)have reached 1.1 or so.However,servers in most of the small and medium-sized data centers are not always in a high load operating state,whose overall utilizations are much lower while the servers still consume a lot of energy when they are idle.Therefore,energy proportional computing has emerged in both industry and academia,and energy proportionality(EP)has been proposed as a metric to measure the relationship of server power consumption and its utilization.Ideally,the power consumption of the server with EP=1.0 should be proportional to its utilization when the server is working.That is,the power consumption of the server with utilization rate of 10% should be one-tenth of its power consumption with utilization rate of 100%.However,servers have different energy proportionality values due to many factors,such as processor process,hardware design,hardware configuration,etc.Virtualization technology has been widely deployed in current data centers for server consolidation,application isolation and flexible management.At the same time,virtualization also saves power and energy consumption due to resource multiplexing and consolidation.In virtualized environment,virtual machine monitor(VMM),or hypervisor act as the equivalent operating systems and they are responsible for resource scheduling and guest operating system provisioning.While server consolidation can reduce data center power consumption to a certain extent,in a multi-tenants cloud computing environment,cloud service providers tend to adopt over provisioning strategy to respond to intermittent burst workloads in order to ensure the quality of service for different customers' virtual machines.Therefore,appropriate scheduling of virtual machines according to its workload characteristics and the energy consumption of different servers and clusters in the data center,can not only provide further server consolidation,but also ensure the quality of service guarantee of tenants as well as reduction of energy consumption in data centers.Thus motivates the work in this thesis.In this thesis we first analyze the published results of the SPECpower benchmark from 2007 to 2016,and make an in-depth study on the energy consumption of real commercial production servers and their energy efficiencies.Our analysis reveals the relationship between energy efficiency,energy proportionality and processor architecture,hardware configuration,and server performance.Moreover,for servers with higher energy proportionality value(EP> 0.8),the non-linearity of energy consumption corresponding to its performance is becoming more and more dominant.At the same time,the energy efficiency curve of the server is also nonlinear.Most importantly,many newly manufactured servers tend to achieve its peak energy efficiency point at its non-100% utilization point.Therefore,through the virtual machine scheduling and migration to keep the cluster server running at the peak efficiency of each node,will be able to improve the overall energy efficiency of the cluster.In this thesis we perform an extensive running a computing intensive,hybrid workload,a comparison between the current mainstream cloud service provider virtualization platform,Microsoft Hyper-V,VMware ESXi,Xen,KVM,and the container virtualization engine,Docker,to reveal their energy efficiency characteristic.According to the energy consumption characteristics of the server and the energy efficiency characteristics of the virtualized platform,this thesis proposes an energy efficiency-aware virtual machine scheduling policy to improve the overall energy efficiency of the server cluster and to reduce its power consumption.The experimental results show that the power consumption can be saved about 37.07% ~ 49.98% in the homogeneous node cluster,and the average completion time of the computing intensive jobs increases only 0.31% ~ 8.49%,while in the heterogeneous nodes,the power consumption of the computing intensive jobs can be reduced by 44.22% and the job completion time can be saved by 53.80% by using the scheduling algorithm proposed in this thesis.The results presented in this thesis provide useful insights for the power and energy management of virtualized data centers containing heterogeneous node server clusters.At the same time,the energy efficiency aware virtual machine scheduling policy proposed in this thesis can also serve as a good reference for green data center operation based on server energy efficiency,energy proportionality and workload characteristics.