SHF: Small: Green Parallel Language Systems

点击翻译按钮获取中文摘要

Title: SHF:Small:Green Parallel Language SystemsThe broad spectrum of modern parallel computing platforms are faced with the growing concern about energy consumption. This project explores the frontier at the converging point of parallel computing and energy-efficient computing. The intellectual merits of the proposal are to study energy efficiency of parallel computing through novel programming language and compiler technologies, and produce a set of energy-optimizing compilers, language runtimes, and programming models for parallel systems. With parallel platforms pervasively deployed in the modern computing world, the project's broader significance and importance are as diverse as reducing the operational cost of data centers, improving system reliability of mission-critical systems, extending the battery life of handheld devices, and promoting the environmental sustainability of our society. The unifying theme of the project is to exploit program structures and run-time semantic information for optimizing energy consumption of multi-threaded programs on parallel architectures. Concretely, the project consists of three interconnected directions. (i) energy-efficient thread management, that is, optimizing energy consumption through coordinating threads based on their static and dynamic dependencies. (ii) energy-efficient data management, that is, promoting energy proportionality through analyzing data-intensive programs in a setting where data rates, data processing schedules, and CPU frequencies may interplay in complex fashions. (iii) energy-efficient synchronization management, that is, harvesting different thread synchronization patterns to achieve pattern-specific energy optimization. Together, the three directions constitute a unique and important dimension of energy optimization for parallel computing platforms.