CSR: Small: System Support for Transiency in Data Center and Cloud Computing

今天的分布式应用程序是基于这样一种隐含的假设构建的，即底层数据中心服务器将是稳定且正常可用的，除非偶尔出现故障。然而，在许多新兴场景中，数据中心和云只提供短暂的服务器可用性，而不是持续的可用性。现代分布式系统的暂时性出现在许多环境中，例如使用可再生间歇性资源供电的绿色数据中心、提供可从用户抢占的低成本Spot服务器实例的云平台，或者当智能电网发出信号时自动减少能源使用的智能数据中心。该项目建议在分布式系统设计中进行新的研究，将瞬变作为一流的设计原则。由于传统的容错方法过于昂贵或不适合处理由瞬变引起的服务器间歇性可用，本项目针对现代数据中心和云平台提出了新的低成本瞬变机制。针对绿色数据中心的暂时性，本项目重点研究了一种基于有界迁移的新型虚拟化机制，并将其应用于系统级暂态管理算法的设计。对于云平台，该项目专注于瞬时算法，这些算法可以智能地管理稳定的按需服务器和瞬时Spot服务器的不同组合。最后，为了处理智能电网交互产生的暂态，该项目提出了新的机制，在不影响应用程序性能的情况下明智地调节能源使用。该项目采用基于原型实施和实验评估的系统驱动方法，使用实际应用程序来展示这些过渡机制的好处。包含瞬变的分布式系统的设计有可能显著增加数据中心内可再生能源的使用，并更好地将现代数据中心和云平台与智能电网相结合。该项目的更广泛影响将包括纳入研究成果的研究生课程、本科暑期研究项目、在绿色计算方面与当地K-12学校的接触，以及向研究界发布过渡算法的源代码。

Today's distributed applications are built using the implicit assumption that the underlying data center servers will be stable and normally available, barring for occasional faults. In many emerging scenarios, however, data centers and clouds only provide transient, rather than continuous, availability of their servers. Transiency in modern distributed systems arises in many contexts such as green data centers powered using renewable intermittent sources, cloud platforms that provide lower-cost spot server instances which can be preempted from their users, or smart data centers that voluntarily curtail their energy usage when signaled by the smart electric grid. This project proposes new research in distributed systems design that treats transiency as a first-class design principle. Since traditional fault-tolerance methods are too expensive or not well suited to handling the intermittent availability of servers caused by transiency, this project proposes new low-cost transiency mechanisms for modern data center and cloud platforms. For transiency in green data centers, the project focuses on a new virtualization mechanism based on bounded migration, and applies it to design system-wide transiency management algorithms. For cloud platforms, the project focuses on transiency algorithms that intelligently manage different mixes of stable on-demand and transient spot servers. Finally, to handle transiency due to smart grid interactions, the project proposes new mechanisms to judiciously modulate energy use without impacting application performance. The project employs a systems-driven approach based on prototype implementation and experimental evaluation using realistic application to demonstrate the benefits of these transiency mechanisms. The design of distributed systems that incorporate transiency have the potential to significantly increase the use of renewable energy sources within data centers and better integrate modern data centers and cloud platforms with the smart grid. Broader impacts of the project will include graduate courses that incorporate research results, undergraduate summer research projects, outreach to local K-12 schools in green computing, and the release of source code for the transiency algorithms to the research community.