A Holistic Approach to Compiler-Assisted Optimization of Software Systems

国家科学基金分布式系统研究CISE/CNSABSTRACTPROPOSAL NUMBER: 0410918主要研究者：Debray， Saumya k .机构：亚利桑那大学提案题目：软件系统编译辅助优化的整体方法本项目旨在通过考虑操作系统和应用程序以整体方式一起运行在处理器上，并考虑这些系统的整体和个体行为来改进计算机软件。计算机系统的每个组成部分，即操作系统和不同的应用程序，都在自己的地址空间中运行。现有的软件优化方法将这些组件视为孤立的实体，并且不考虑其他地址空间的行为或与其他地址空间的交互。这样做的缺点是软件改进过程不能利用不同地址空间中代码模块之间的交互知识。本项目旨在开发工具和技术来解决这一缺陷。我们方法中的一个关键元素是以新颖的方式使用编译器技术，特别是在多地址空间系统上下文中，而不是在单地址空间应用程序级上下文中。这项工作的影响将是根据几个不同的度量标准，在各种各样的环境中提高软件系统的性能。特别是，我们希望能够改进用户应用程序、操作系统内核和系统服务，性能指标包括执行速度、所需内存量、容错性和增强的安全性。我们的结果将以出版物的形式传播，描述我们的算法、实验方法和结果，以及作为项目一部分开发的软件，我们将通过全球网络提供。Brett D. fleisch CISE/ cns项目主任2004年6月24日。

National Science FoundationDistributed Systems Research CISE/CNSABSTRACTPROPOSAL NUMBER: 0410918PRINCIPAL INVESTIGATOR: Debray, Saumya K.INSTITUTION: U of Arizona PROPOSAL TITLE: A Holistic Approach to Compiler-Assisted Optimization of Software Systems This project aims to improve computer software by considering the operating system and application programs running on a processor together in a holistic manner, and consider-ing global as well as individual behaviors of such systems. Each component of a com-puter system, i.e., the operating system and the different applications, runs in its own ad-dress space. Existing approaches to software optimization consider these components as isolated entities and assume nothing about the behavior of, or interactions with, other ad-dress spaces. This has the shortcoming that the software improvement process cannot take advantage of knowledge of interactions between code modules in the different ad-dress spaces. This project aims to develop tools and techniques to address this shortcom-ing. A key element in our approach is the use of compiler techniques in novel ways, and in particular, in a multi-address space systems context rather than a single address space application-level context.The impact of this work will be to improve the performance of software systems across a wide variety of contexts and according to several different metrics. In particular, we ex-pect to be able to improve user applications, operating systems kernels, and system ser-vices, with performance metrics that include execution speed, amount of memory re-quired, fault tolerance, and enhanced security. Our results will be disseminated in the form of publications describing our algorithms, experimental methodology and results, as well as software developed as part of the project, which we will make available via the World-wide web.Dr. Brett D. FleischProgram Director, CISE/CNSJune 24, 2004..