Energy Accounting And System Optimization In Cloud Datacenter

The datacenter is the basic support facility for cloud computing.In recent years,thanks to the promotion of cloud computing,the scale of the datacenter has expanded and deployed widely all over the world.In the meantime,the energy consumption of the datacenter has increased sharply as well.On the one hand,the rising energy consumption has increased the operating costs of datacenters.On the other hand,datacenters are under pressure from governments and environmental organizations because of a large amount of carbon emissions.Under the background of the fierce market competition of cloud computing and the increasing environmental pressure,how to reduce the energy cost of datacenters and carbon emissions has already caused widely concern from industry and academia.Energy accounting is the bedrock of the datacenter energy management.A reasonable energy metering method is conducive for the datacenter to analyze energy consumption costs,come up with a reasonable pricing strategy and provide the basis for energy efficiency optimization.In a datacenter,the energy consumption is mainly generated by IT and non-IT facilities.For the purpose of improving utilization,datacenters usually use virtualization technology to manage IT resources,which makes virtual machines the basic unit of resource management of datacenters.However,as a logical abstraction of physical hardware resources,virtual machines are not able to measure its energy consumption directly through hardware.Moreover,another problem is that resource competition between virtual machines will lead to changes in energy consumption of virtual machines,while the resource-to-energy mapping model used in the current research cannot effectively reflect the impact of resource competition on energy consumption.On the other hand,non-IT facilities also account for a large proportion of datacenter energy consumption,and their energy consumption and efficiency are greatly affected by the load of IT facilities.The research on energy metering and optimization in this part is still not enough currently.In addition,it is found that the IT load changes greatly during the actual operation of the datacenter,and is far below the design peak,resulting in low utilization and low energy efficiency of the corresponding non-IT facilities.To solve the above problems,two efficient and fair energy accounting methods are designed to provide fine-grained IT and non-IT energy of each virtual machine.In addition,a hybrid water cooling system is designed to overcome the drawbacks caused by varied IT load and improve the datacenter energy efficiency.Specifically,it includes the following three aspects:In order to solve the problem that energy consumption change caused by virtual machine resource competition is difficult to quantify,a new criterion consisting of macro-level accuracy and fairness is designed based on game theory.Based the new criterion,a fair virtual machine energy accounting method is proposed by combining the running state of the virtual machine with the Shapley value method in the economic methodology.The experimental results show that the average error of the proposed method is less than 6%and the maximum error is 11.7%.To make the Shapley value more suitable for a large scale datacneter,a lightweight energy accounting method based on the Shapley value is developed.The computational complexity of the Shapley value is up to O(2~N),while a non-IT facility typically serves thousands of IT devices.Hence,the Shapley value method is hard to apply directly to the measurement of non-IT energy consumption.By measuring and analyzing the energy consumption of non-IT facilities in real datacenters and deriving on the basis of the energy consumption characteristics of non-IT facilities,a method equivalent to the Shapley value method but with a complexity of only O(N)is designed.The experimental results in real-world deployment show that the average error the proposed method is less than 4%and the maximum error is 6.97%.To improve the energy usage effectiveness in datacenters,an energy-efficient hybrid water cooling system is designed.The low utilization of servers leads to a large amount of redundant cooling capacity,resulting in energy waste.Due to technical limitations,current water cooling systems generally use centralized cooling control methods,and its energy efficiency also suffers from the utilization imbalance among servers.The new hybrid water cooling system combines water cooling and thermoelectric coolers,and designs a fine-grained cooling control mechanism based on IT load changes to improve the cooling energy efficiency of the datacenter with a partial PUE of 1.04?1.05.The design of the above two energy accounting methods provides the datacenter a comprehensive and fine-grained perspective on the energy consumption,and the IT load-aware hybrid water cooling system improves the energy efficiency of the datacenter.