Use of operating system scheduling for enhanced performance and usability of chip multiprocessing systems

使用操作系统调度来增强芯片多处理系统的性能和可用性

• 批准号: 346192-2007
• 负责人: Fedorova, Alexandra
• 金额: $ 1.42万
• 依托单位: Simon Fraser University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2007
• 资助国家: 加拿大
• 起止时间: 2007-01-01 至 2008-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=365982
• 关键词: Use; operating; system; scheduling; enhanced

This project will investigate operating system scheduling algorithms for concurrent commercial applications and multi-instance virtual machine (VM) workloads on chip multiprocessors.Chip multiprocessing (CMP) systems are emerging as popular computing platforms due to their superior performance and power efficiency. Unlike conventional processors, chip multiprocessors employ fine-grained sharing of their internal resources. They run multiple tasks (i.e., processes or threads) on the same physical chip, and the tasks share the cache and other processor's components. Operating system scheduling - that is, the assignment of tasks to processors - has been shown effective at producing performance improvements on chip multiprocessors, because a scheduler defines how tasks share the processor's internal resources. Operating system scheduling is an attractive vehicle, because changes to the operating system software can be deployed more quickly and at lower expense than any solution that requires changes to the hardware.This project will investigate scheduling algorithms for two important workload types that were not studied extensively in the past: concurrent commercial applications and multi-instance VM workloads. These workloads will be used much on CMP systems, because their inherent parallelism dovetails with the parallel design of CMP hardware. We will design, implement and evaluate scheduling algorithms that target CMP systems' performance and usability for these workloads.Our initial research will address adaptations to existing operating systems, assuming that we cannot change the hardware or applications.  However, in the long run we hope to gain insight into how all system components can be adapted or redesigned to assure the most productive functioning of CMP systems.

本项目将研究在片上多处理器上的并发商业应用和多实例虚拟机（VM）工作负载的操作系统调度算法。片上多处理（CMP）系统由于其上级性能和功率效率而成为流行的计算平台。与传统的处理器不同，片上多处理器采用细粒度的内部资源共享。它们运行多个任务（即，进程或线程），并且任务共享该高速缓存和其他处理器的组件。操作系统调度-即任务分配给处理器-已被证明有效地提高了芯片多处理器的性能，因为调度程序定义了任务如何共享处理器的内部资源。操作系统调度是一个有吸引力的工具，因为对操作系统软件的更改可以比任何需要更改硬件的解决方案更快地部署，成本更低。本项目将研究过去没有广泛研究的两种重要工作负载类型的调度算法：并发商业应用程序和多实例VM工作负载。这些工作负载将在CMP系统上大量使用，因为它们固有的并行性与CMP硬件的并行设计相一致。我们将针对CMP系统的性能和可用性设计、实现和评估调度算法。我们的初步研究将解决对现有操作系统的适应性问题，假设我们不能改变硬件或应用程序。然而，从长远来看，我们希望深入了解如何调整或重新设计所有系统组件，以确保CMP系统的最高生产力功能。