Towards Energy Efficient Cloud Data Centers - A Framework for Evaluation and Analysis of Energy Efficiency

Abstract

The purpose of the research questions proposed in the thesis were to investigate what methods and combinations of methods could illustrate areas of energy inefficiency at the analyzed and evaluated cloud data centers. However, also investigating how the methods utilized effect other data center devices monitored. With the purpose of the research question lastly resulting in an illustration of areas of energy inefficiency and what actions a cloud provider could take to reduce energy consumption to reduce environmental impact and operational cost. The project was scoped to the targeted areas compute, storage, and network. with the compute category targeting the data center physical hosts, while storage was scoped to the data center disks, and for the networked scoped to switch devices. With areas such as the data center cooling being left out of the scope of the project due to it's complexity and time requirements on top of compute, storage, and networking. With the work done in the thesis being exploration of prior work and background information regarding energy efficiency in data centers, with the work further developing an analysis and evaluation framework. Where the resulting framework could analyze a cloud provider's cloud data centers to provide insight into resource utilization and placement of workloads. In addition to further evaluating the cloud provider's cloud data center by estimating the reduction of energy consumption from different methods. As a result of the work it was discovered that not only did the framework proposed methods which could potentially reduce power consumption at cloud data center and thus energy consumption. The framework also through the provided insight could inform the user of actions that could improve the energy efficiency through improved compatibility between workloads and devices. Where the placement methods utilized led to greater energy efficiency for the monitored devices with reorganizing placement approach which in addition takes into account the over utilization led to in the case of the physical hosts an increased energy consumption at eight point forty-two as supposed to zero point zero two percent reduction in energy consumption. while the effected data center network reduced energy consumption by thirty-three point eighty-five percent and twenty-nine point forty-four percent. With the analysis and evaluation also illustrating a lack of available memory resources as the workloads utilize consumed more RAM than CPU. The thesis thus concluded by summarizing the work done, but also suggesting improvements to the framework such as the implementing dynamic power scaling approaches to data center disks and network links. Additionally did the thesis also propose future work such as testing the solution in a real data center environment and implementing analysis and evaluation for the data center cooling.