Constrained Power and Performance Optimization for Embedded Systems

嵌入式系统的受限功率和性能优化本研究的目标是探索嵌入式系统的电源管理技术，这些技术具有可证明的功率效率界限，以及由于电源管理对延迟的不利影响。系统级电源管理的当前状态仅限于在一段空闲时间后关闭系统的某些部分，从而忽略了应用程序时间限制、运行时流量和应用程序使用信息。这项工作将开发电源性能控制“旋钮”，使我们能够做出有效的在线决策。具体的应用将包括RTOS中的电力感知资源调度，时间感知的电力优化以及电力节约和应用程序服务质量（例如，错过截止日期）之间的权衡。我们的技术重点是解决两个关键问题：(a)延迟受限的功率优化，即最小化系统级功耗，同时限制电源管理导致的系统延迟的影响；(b)受功率限制的性能优化，例如，在给定的功率预算内实现系统级任务和调度。作为第一步，我们通过开发由于电源管理导致的延迟增加的界限，专注于在线电源管理算法的有效性和效率的分析界限。我们引入竞争比率的概念，作为给定电源管理算法在最佳功耗配置文件下的性能的定量度量。接下来，我们将这些分析边界合并到更广泛的系统级时序和功率仿真引擎中，从而实现准确的性能仿真，同时最大限度地减少与实际系统功能相关的细节。我们的实验评估是通过RTOS通过协调调度和资源关闭实现新的电源管理服务。

Constrained Power and Performance Optimization for Embedded SystemsThe goal of this research is to explore techniques for power management ofembedded systems with provable bounds on power efficiency as well as adverseeffects on latency due to power management. The current state of the art insystem-level power management is limited to shutting parts of a system aftera certain period of idle time, thus ignoring the application timingconstraints, runtime traffic and application usage information. This workwill develop power-performance control "knobs" that will allow us to makeeffective online decisions. Specific applications will include power-awareresource scheduling in RTOS, timing-aware power optimizations and tradeoffsbetween power savings and application quality of service (e.g., misseddeadlines).Our technical focus is on solving two key problems: (a) latency-constrainedpower optimization, i.e., minimization of system-level power consumptionwith constrains on the effect of system latency due to power management; and(b) power-constrained performance optimization, e.g., system-level taskimplementation and scheduling within a given power budget. As a first step,we focus on analytic bounds on the effectiveness of online power managementalgorithms and their efficiency by developing bounds on latency increasesdue to power management. We introduce the notion of a competitive ratio asa quantitative measure of how well a given power management algorithmperforms against an optimum power consumption profile. Next, we incorporatethese analytic bounds in a broader system-level timing and power simulationengine which enables an accurate performance simulation while minimizing thedetails related to actual system functionality. Our experimental evaluationis through RTOS implementation of new power management services throughcoordinated scheduling and resource shutdown.