Chorus: A Resource Management Framework for Cost Effective Quality of Service in Cloud Environments

随着互联网上经济和协作交换的全部潜力变得清晰，基于web的应用程序和用户需求呈指数级增长。许多服务现在依赖于共享的基础设施，称为云环境，目前由大型服务托管提供商维护。不幸的是，在幕后，云提供商被大量的管理成本所束缚，这些成本消耗了高达88%的预算，主要是由于服务器管理、电源和冷却操作成本。这限制了再投资、研究和发展，从而阻碍了该行业的效率，而且由于主要的能源消耗，也造成了环境问题。对灵活、无处不在、可定制的云访问的需求只会以急剧的速度增长。

As the full potential of economic and collaborative exchange over the Internet has become clear, Web-based applications and user demand have increased exponentially. Many services now rely on shared infrastructures, called Cloud environments, currently maintained by large service hosting providers. Unfortunately, behind the scenes, Cloud providers are strapped with large administrator costs, costs which consume up to 88% of their budgets, mainly due to server management, power and cooling operational costs. This creates an impediment on the efficiency of this industry, by limiting reinvestment, research and development, and also an environmental problem due to major energy consumption. And demand for flexible, ubiquitous, customizable Cloud access is only going to increase at a steep rate. To address these problems, we propose to design, implement and deploy a novel Cloud resource management framework, Chorus, intended for use as a collaborative environment for interdisciplinary research. Chorus is a lightweight and portable virtualization solution including algorithms, high level languages and tools for performance modeling, anomaly diagnosis and efficient resource allocation. Our techniques will allow for automatic detection and minimization of application interference for resources in Cloud service hosting environments, and automatic power savings, as well as for flexible what-if inquiry for capacity planning and anomaly diagnosis by the system administrator. By these novel proposed techniques, our system, Chorus, will automatically provide performance optimizations towards improving end-to-end Quality of Service compliance for all hosted applications as well as cost reductions for the hosting provider. We will demonstrate the capabilities of our in-lab Cloud environment by running a variety of realistic workloads, ranging from common e-commerce workload mixes modeled after the workloads of the Amazon.com and EBay.com, to functional MRI medical image processing applications.