ITR: Interactions of Multithreaded Computing and Operating System Design

提案编号：0085670标题：Interaction of Multithreaded Computing and Operating System DesignPI：Henry M.Levy和Susan J.Eggers摘要：在接下来的十年中，由数百或数千个节点组成的高度可伸缩的计算机系统将成为满足新兴的全球应用的需求所必需的，例如用于因特网商业的Web和数据库服务器。这些应用程序必须双线程化才能处理数千个并发请求。这个可伸缩服务器域的一项很有前途的技术是多线程处理器的使用；例如，Compaq已经宣布在2002/2003年的时间框架内对其Alpha处理器提供同步多线程(SMT)支持。不幸的是，还没有关于操作系统的多线程支持或如何为多线程CPU构建操作系统的研究。然而，操作系统在这种环境中可能是至关重要的；例如，测量表明，加载的Web服务器可以将其75%的时间花在操作系统上。所提出的研究将开发软件环境，以补充SMT和其他细粒度并行处理器提供的多线程硬件环境。研究直接位于体系结构和操作系统之间，检查(1)支持操作系统需求的多线程处理器的设计和性能，以及(2)根据多线程处理器的能力来检查操作系统的结构。这些研究将集中在高度并行的、请求驱动的工作负载领域，如Web服务器和数据库服务器。

Proposal Number: 0085670Title: Interactions of Multithreaded Computing and Operating System DesignPIs: Henry M. Levy and Susan J. EggersAbstract:Over the next decade, highly scalable computer systems consisting ofhundreds or thousands of nodes will become necessary to meet thedemands of emerging global applications, such as Web and databaseservers used for Internet commerce. These applications must bemultithreaded to handle the thousands of simultaneous requests. Onepromising technology for this scalable server domain is the use ofmultithreaded processors; for example, Compaq has announcedSimultaneous Multithreading (SMT) support for its Alpha processor in the 2002/2003 time frame.Unfortunately, no research has been conducted on multithreadingsupport for operating systems or on how to structure operating systemsfor multithreaded CPUs. Yet the OS may be crucial in this environment;e.g., measurements show that a loaded web server can spend 75% of itstime in the OS. The proposed research will develop the softwareenvironment to complement the multithreaded hardware environmentprovided by SMT and other fine-grained parallel processors. Theresearch sits squarely between architecture and operating systems,examining (1) the design and performance of multithreaded processorsto support OS needs, and (2) the structure of operating systems inlight of the capabilities of multithreaded processors. The researchwill focus on these questions in the domain of highly parallel,request-driven workloads, such as Web servers and database servers.