Foundations for Heterogeneous Datacenter Design and Management
撰稿：
摄影：
发布时间：2013年03月07日
时间：2013年3月8日（周五）下午3:00-4:30
地点：446会议室
摘要 As cloud computing proliferates, demand for datacenter computing capacity increases. Moreover, we must increase capacity within today's megawatt-scale power budgets. If we could achieve such datacenter architectures, we would see qualitative advances in diverse application domains, such as computational science, business analytics, and personal computing. Toward this goal, we present the case for building datacenters using processors and memories that were originally intended for mobile and embedded platforms.
Mobile processors reduce power by simplifying the datapath and cache. For web search, mobile processors are 5x more efficient than server processors. We quantify and mitigate the impact on query latency, relevance, and quality-of-service. Mobile memories reduce power by using non-terminated links and lower data rates. Such low-power memory is 5.6x more efficient than server memories. We identify datacenter applications that can benefit from mobile memories. Finally, diverse applications have heterogeneous architecture demands and we discuss how future datacenters might navigate this complexity with statistical inference and economic mechanisms.
主讲人简介 Benjamin Lee is an assistant professor of Electrical and Computer Engineering at Duke University. His research focuses on scalable technologies, power-efficient architectures, and high-performance applications. He is also interested in the economics and public policy of computation. He has held visiting research positions at Microsoft Research, Intel Labs, and Lawrence Livermore National Lab.
Dr. Lee received his B.S. at the University of California at Berkeley, S.M. and Ph.D. at Harvard University, and post-doctorate at Stanford University. He received the NSF CAREER Award in 2012. And his research has been honored as a Top Pick by IEEE Micro Magazine (2010), twice as a Research Highlight by Communications of the ACM (2010, 2011), and by an NSF Computing Innovation Fellowship (2009-10).
附件下载：
Foundations for Heterogeneous Datacenter Design and Management
地点：446会议室
摘要
As cloud computing proliferates, demand for datacenter computing capacity increases. Moreover, we must increase capacity within today's megawatt-scale power budgets. If we could achieve such datacenter architectures, we would see qualitative advances in diverse application domains, such as computational science, business analytics, and personal computing. Toward this goal, we present the case for building datacenters using processors and memories that were originally intended for mobile and embedded platforms.
Mobile processors reduce power by simplifying the datapath and cache. For web search, mobile processors are 5x more efficient than server processors. We quantify and mitigate the impact on query latency, relevance, and quality-of-service. Mobile memories reduce power by using non-terminated links and lower data rates. Such low-power memory is 5.6x more efficient than server memories. We identify datacenter applications that can benefit from mobile memories. Finally, diverse applications have heterogeneous architecture demands and we discuss how future datacenters might navigate this complexity with statistical inference and economic mechanisms.
主讲人简介
Benjamin Lee is an assistant professor of Electrical and Computer Engineering at Duke University. His research focuses on scalable technologies, power-efficient architectures, and high-performance applications. He is also interested in the economics and public policy of computation. He has held visiting research positions at Microsoft Research, Intel Labs, and Lawrence Livermore National Lab.
Dr. Lee received his B.S. at the University of California at Berkeley, S.M. and Ph.D. at Harvard University, and post-doctorate at Stanford University. He received the NSF CAREER Award in 2012. And his research has been honored as a Top Pick by IEEE Micro Magazine (2010), twice as a Research Highlight by Communications of the ACM (2010, 2011), and by an NSF Computing Innovation Fellowship (2009-10).