Resource Scheduling with QoS for Parallel I/O Systems

并行 I/O 系统的具有 QoS 的资源调度

• 批准号: 0541369
• 负责人: Peter Varman
• 金额: --
• 依托单位: William Marsh Rice University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2006
• 资助国家: 美国
• 起止时间: 2006-08-15 至 2012-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0541369&HistoricalAwards=false
• 关键词: Resource; Scheduling; QoS; Parallel; Systems

The architecture and usage modalities of storage systems have been revolutionized by advances in disk drive and networking technologies. Resource management issues occupy a central and increasingly important role in the operation of data centers that must coordinate the operation of large numbers of concurrent devices and serve hundreds of gigabytes of data per second. Such shared storage servers are being proposed as a cost-effective solution for maintaining data repositories. This research effort addresses resource allocation and scheduling issues that arise in multiplexing a shared storage server and seeks to provide isolation and performance guarantees for multiple contending flows that may have different characteristics. The research is developing robust models, algorithms, and mechanisms that enable efficient sharing of server I/O resources while enforcing performance guarantees, in the presence of dynamically varying resource constraints. The results will also have applicability to systems of heterogeneous servers. The intellectual advance pursued in this research is the systematic investigation of a new set of resource management issues that arise in the shared server environment. It studies, at a fundamental level, the problem of providing fair or differentiated quality of service among diverse concurrent clients (flows) accessing a shared parallel I/O system. The work is developing robust models that incorporate the unique resource constraints and performance responses arising in parallel I/O, developing efficient algorithms for adaptive resource allocation, formal characterization of the solutions with respect to performance and stability, and the development of simulation infrastructure for empirical evaluation of the solutions.The impact of the research will be more robust and scalable resource management for data centers such as commercial and supercomputing server environments. The research will produce mechanisms to enforce service agreements between the service provider and concurrent clients, reaping the economies of scale of shared resources and the increasing economic benefits of consolidated management. The research will also result in the training and education of graduate students in a growing sector of the information technology industry.

磁盘驱动器和网络技术的进步已经彻底改变了存储系统的架构和使用方式。资源管理问题在数据中心的运营中占据着越来越重要的中心地位，数据中心必须协调大量并发设备的运营并每秒提供数百GB的数据。此类共享存储服务器被提议作为维护数据存储库的经济有效的解决方案。这项研究工作解决了多路复用共享存储服务器时出现的资源分配和调度问题，并寻求为可能具有不同特征的多个竞争流提供隔离和性能保证。该研究正在开发稳健的模型、算法和机制，在存在动态变化的资源约束的情况下，能够有效共享服务器 I/O 资源，同时强制执行性能保证。结果也适用于异构服务器系统。本研究追求的智力进步是对共享服务器环境中出现的一组新资源管理问题的系统研究。它从根本上研究在访问共享并行 I/O 系统的不同并发客户端（流）之间提供公平或差异化服务质量的问题。这项工作正在开发鲁棒模型，其中包含并行 I/O 中出现的独特资源约束和性能响应，开发用于自适应资源分配的高效算法，对性能和稳定性解决方案进行正式表征，以及开发经验仿真基础设施该研究的影响将为数据中心（例如商业和超级计算服务器环境）提供更强大和可扩展的资源管理。 该研究将建立机制来执行服务提供商和并发客户之间的服务协议，从而获得共享资源的规模经济和综合管理不断增加的经济效益。该研究还将为信息技术行业不断发展的领域的研究生提供培训和教育。