CISE Research Infrastructure: Infrastructure for Power-Aware Server Clusters

EIA 02-24428 Bianchini，RicardKremer，UlrichRutgers University New Brunswick标题：CEISE RR：节能服务器集群的基础设施这项提案调查了集群环境中涉及的电力、能源、性能和功能问题以及权衡，旨在获得用于研究节能服务器集群的设备。基础设施的主体由一群“刀片”服务器组成。由于刀片式服务器比传统服务器消耗更少的电力，并允许更精细地控制电力状态，因此它们被认为是进行电力和能源节约研究的理想之选。该基础设施支持两项研究：基于集群的互联网服务器的节能策略，如大容量WWW、数据库或文件服务器；防篡改和容错传感器网络环境下的计算和数据服务器。前者涉及支持大多数WWW数据/托管中心、互联网公司和大量教学和研究机构的服务器，但消耗大量的电力和能源。节约这些资源对于降低这些群集的安装和运行成本至关重要。在不降低性能的情况下(或至少在不过度降级的情况下)节约能源和能源构成了挑战。建议的集群刀片能够设计和试验能够调度资源需求的技术，允许从活动模式过渡到低功率模式，而不会造成相关的性能损失。该项目使用不同的服务器进行实验，其中包括刀片式服务器和传统的耗电服务器。后一个项目设计了传感器网络，以识别、监控和跟踪物理空间内的对象或主体。整个空间被划分为子空间，每个子空间都有自己的服务器(或服务器的小集群)，所有服务器都通过网络连接。该设计允许对传感器数据进行局部和全局处理和评估。同样，能源效率在这种情况下变得至关重要，因为服务器需要能够承受较长时间的电源故障，并仍然提供关键服务。刀片式服务器集群能够研究此环境中关键服务器任务的能源行为，并试验可在不显著降低其性能和功能的情况下降低这些任务的能源消耗的技术。在教育方面，刀片式服务器和系统将用于开发操作系统、分布式计算和编译器课程。推广电力和能源消耗以及性能、设计、实施和评估指南。

EIA 02-24428Bianchini, RicardKremer, UlrichRutgers University New Brunswick Title: CISE RR: Infrastructure for Power-Aware Server Clusters This proposal, investigating the power, energy, performance, and functionality issues and trade-offs involved in cluster environments, aims at acquiring equipment for research on power aware clusters of servers. The bulk of the infrastructure consists of a cluster of "blade" servers. Since blade servers consume less power and allow finer control of power states than traditional servers, they are considered ideal for power and energy conservation research. The infrastructure supports two research studies in:Strategies for conserving power and energy in cluster-based Internet servers, such as high-capacity WWW, database, or file serversComputation and data servers in the context of a tamper-resistant and fault-tolerant sensor network.The former project involves servers that support most WWW data/hosting centers, Internet companies, and a large number of teaching and research institutions, but consume significant amounts of power and energy. Conserving these resources is critical to reduce the installation and operation costs of these clusters. Conserving power and energy without degrading performance (or at least without degrading excessively) constitutes the challenge. The cluster blades proposed enable designs and experiments with techniques that can schedule the demand for resources, permitting transitioning from active to low-power model without relevant performance penalty. The project experiments with heterogeneous servers comprising combination of blades and traditional, power-hungry servers. The latter project designs the sensor network to identify, monitor, and track objects or subjects within a physical space. This overall space is divided into subspaces, where each subspace has its own server (or small cluster of servers), and all servers are connected via a network. The design allows local and global processing and evaluation of sensor data. Again, energy efficiency becomes crucial in this context because the servers need to be able to survive power failures for extended periods of time and still provide critical services. The cluster of blades enables the study of energy behavior of the critical server tasks in this environment, as well as experiment with techniques that can reduce the energy consumption of these tasks without noticeably degrading their performance and functionality.On the educational side, the blades and the systems will be exploited developing operating systems, distributed computing, and compiler courses. Power and energy consumption, as well as performance, design, implementation, and evaluation guidelines will be promoted.