Understanding and Improving Memory System Performance via a New Approach to Cache Analysis and Access Ordering

标题：通过高速缓存分析和访问排序的新方法了解和改进内存系统性能PI：犹他大学的Sally McKee微处理器速度的增长速度远远快于内存系统速度：这两条速度-增长曲线都是指数型的，但它们代表发散指数。这项研究致力于开发一种全面的方法来弥合处理器/内存性能差距，方法是开发更好的性能度量，更好地观察内存系统行为和理解内存使用模式的工具，以及结合编译器和体系结构方案来利用这些模式。本研究的第一部分研究了更好地测量和解释高速缓存和内存层次中的复杂系统行为的方法和工具。PI和其他人正在开发一个新的缓存系统分析框架；这个新的“缓存即过滤器”框架关注的是内存跟踪的属性，而不是特定内存组织的构件。这项研究将支持将新框架应用于实际问题和系统所需的工具和技术的开发。本研究的第二部分研究了结合编译器和体系结构的方法，通过更智能地调度内存活动和硬件对流的支持来提高现代微处理器中的内存系统性能。

ABSTRACTTitle: Understanding and Improving Memory System Performance via a New Approach to Cache Analysis and Access orderingPI: Sally McKee, University of UtahMicroprocessor speed is increasing much faster than memory system speed: both speed-growth curves are exponential, but they represent diverging exponentials. This proposed research strives to develop a comprehensive approach to bridging the processor/memory performance gap by developing better measures of performance, better tools for observing memory system behavior and understanding memory usage patterns, and combined compiler and architectural schemes to exploit these patterns.The first component of this research investigates methods and tools by which to better measure and interpret complex system behavior in the cache and memory hierarchy. The PI and others are developing a new framework for the analysis of caching systems; this new "caches as filters" framework focuses on properties of a memory trace rather than artifacts of a particular memory organization. This research will support the development of the tools and technology necessary to apply the new framework to real problems and systems.The second component of this research investigates combined compiler and architectural approaches to improving memory system performance in modern microprocessors via smarter scheduling of memory activity and hardware support for streaming.