PCP: A Generalized Approach to Optimizing Performance Under Power Constraints through Resource Management
PCP: A Generalized Approach to Optimizing Performance Under Power Constraints through Resource Management
复制标题
PCP：通过资源管理在功率约束下优化性能的通用方法
DOI：
--
复制
发表时间：
2014
期刊：
影响因子：
--
通讯作者：
M. Maggio
中科院分区：
文献类型：
--
作者：
H. Hoffmann;M. Maggio
许多计算系统受到功率预算的约束。虽然它们可以暂时吸收更多的电力，但这样做会造成不可持续的温度和不必要的电力消耗。开发在功率预算内运行的系统是一个约束优化问题：配置系统内的组件以最大限度地提高性能，同时保持可持续的功耗。这是一个具有挑战性的问题，因为系统中的许多不同组件会影响功率/性能权衡，并且它们以复杂的方式相互作用。先前的方法通过固定一组组件并设计仅管理这一组组件的功率预算框架来解决这些挑战。如果有新的组件可用，那么这个框架必须重新设计和重新实现。本文介绍了PCP，电源预算问题的一般解决方案，适用于任意组的组件，即使他们不知道在设计时或在运行时的变化。为了演示PCP，我们在软件中实现它，并将其部署在Linux/x86平台上。（减）
Many computing systems are constrained by power budgets. While they could temporarily draw more power, doing so creates unsustainable temperatures and unwanted electricity consumption. Developing systems that operate within power budgets is a constrained optimization problem: configuring the components within the system to maximize performance while maintaining sustainable power consumption. This is a challenging problem because many different components within a system affect power/performance tradeoffs and they interact in complex ways. Prior approaches address these challenges by fixing a set of components and designing a power budgeting framework that manages only that one set of components. If new components become available, then this framework must be redesigned and reimplemented. This paper presents PCP, a general solution to the power budgeting problem that works with arbitrary sets of components, even if they are not known at design time or change during runtime. To demonstrate PCP we implement it in software and deploy it on a Linux/x86 platform. (Less)