II-New: RIVER: A Research Infrastructure to Explore Volatility, Energy-Efficiency, and Resilience

点击翻译按钮获取中文摘要

With the end of Moore's Law, all computing systems (smart phones, watches, laptops, and even the cloud) of the future must intelligently manage energy, parallelism, and even trade reliability and lifetime for increased computing performance and storage capacity. This daunting task involves decision making in real-time (milliseconds), in a system with many devices (billions of transistors), and very complex behavior (100 million line software) with consequences for performance or lifetime as much as 100-fold. The Research Infrastructure to explore Volatility, Energy-efficiency, and Resilience (RIVER) at the University of Chicago will create a scientific vehicle that enables research on this critical management problem for applications that underlie extraordinary information access and intelligent behavior we depend on in today's information systems.The RIVER project will design and create an observable, controllable computing system (a research infrastructure) where reliability, lifetime, power, volatility, parallelism, and performance can be varied across as wide a range as 10,000-fold. Under flexible software control, the system will support power capping, clock speed scaling, and enable deep customization of solid-state storage. All of these properties will be supported with fine-grained control and complemented by extensive detailed monitoring. Research results will involve new ways to control and intelligently manage computing performance and lifetime with quantitative improvements as large as 100-fold; such large change may be breakthrough capabilities for those applications in the new reality of hardware technology. The RIVER project will advance computer systems, software, algorithms, and applications research. Because of the broad societal importance of computing, the potential impact of these core capabilities could touch many aspects of government, education, and commerce. The RIVER system and tools will be used in undergraduate and graduate classes, thus exposing them to realities of future computing, and will be shared with other universities to maximize impact.