CSR: Large: Collaborative Research: Multi-core Applications Modeling Infrastructure (MAMI)

该奖项是根据2009年《美国复苏和再投资法案》(公法111-5)提供资金的。高端计算机体系结构中蓬勃发展的革命对性能测量、建模和优化工具的软件基础设施具有深远的影响，这些工具对于提高计算科学的生产力在过去十年中是不可或缺的。问题的核心是，新的多核处理器是下一代系统的基础，从工作组集群到千万亿级的超级计算机。芯片制造商转向多核处理器的主要动机是比传统的单核处理器具有更好的性能功耗比。因此，多核处理器并不等同于传统工具所处理的多个CPU。虽然在了解多核的性能权衡方面正在进行大量工作，但这些工作中的大部分都是临时的，需要一个统一的框架，社区可以以系统的方式为其做出贡献。此外，在理解大规模应用的超级计算机系统中的性能-功率权衡方面所做的工作很少。在多核系统中发生的大量资源共享的背景下，了解性能和性能-功率权衡非常重要。本提案的重点是开发多核应用程序建模基础设施(MAI)，该基础设施将促进对多核系统中的性能、功耗和性能-功耗权衡的系统测量、建模和预测。除了开发Mami之外，拟议的工作还将使用Mami对多核系统上的关键基准测试和应用程序的性能和功耗进行建模、分析和优化。

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5). The burgeoning revolution in high-end computer architecture has far reaching implications for the software infrastructure of tools for performance measurement, modeling, and optimization, which has been indispensable to improved productivity in computational science over the past decade. The heart of the problem is that new multicore processors are the foundation of next generation systems, ranging from workgroup clusters to petascale supercomputers. The main motivation by chip manufacturers for the movement to multicore processors is better performance per watt than the traditional single core processor. Hence, multicore processors are not equivalent to multiple CPUs that traditional tools addressed. While significant work is underway on understanding performance tradeoffs with multicores, much of this work is ad hoc and needs a unifying framework to which the community can contribute in a systematic manner. Furthermore, little work has been done on understanding performance-power tradeoffs in supercomputer systems for large-scale applications. It is important to understand performance and performance-power tradeoffs in the context of the significant resource sharing that occurs in multicore systems. This proposal is focused on developing the Multicore Application Modeling Infrastructure (MAMI) that will facilitate systematic measurement, modeling, and prediction of performance, power consumption and performance-power tradeoffs in multicore systems. In addition to developing MAMI, the proposed work will use MAMI to model, analyze and optimize performance and power consumption of key benchmarks and applications on multicore systems.