The Way of the great learning involves manifesting virtue, renovating the people, and abiding by the highest good.

2008年12月26日星期五

a colloquium talk on challenges of multicore processors

发信人: Windtalking (Old Wolf), 信区: CSArch标 题: a colloquium talk on challenges of multicore processors.发信站: 水木社区 (Tue Dec 9 22:06:38 2008), 站内[Time]: 10:00-12:00 AM, Dec. 15(next monday). [Place]: FIT 1-315, Tsinghua Univ. [Title]: Research issues and Challenges to Advance System Software for Data-intensive Applications in Multicore Processors Xiaodong Zhang Department of Computer Science and Engineering The Ohio State University AbstractCompared with rapid technology advancements in multicore processors andrapid changes from computing-intensive to highly data-intensiveapplications, operating systems have been evolved very slowly forseveral decades. Application users are facing to two major challengesin today's computing environment. On the top level of the systemhierarchy, private and shared caches are equipped for many cores toaccess concurrently, inevitably causing access conflicts to degradeexecution performance. On the bottom level, the performance bottleneckof "memory wall" has been shifted to "disk wall" that is a seriousbottleneck for many data-intensive applications. Since processorcaches and disk storage are not in the major scope of operating systemmanagement, and their increasingly complex operations are nottransparent to application users, the above mentioned performanceissues have not been effectively addressed at any level of computersystems. We have made a continuous effort to enhance operating systems with twoobjectives: (1) to well utilize rich but complex resources of multicoreprocessors and (2) to access disk data as fast as possible. At themulticore processor level, we are developing new resource allocationmanagement to improve the effective caching capacity per core and/orper thread, and to minimize congestion in off-chip memory accesses bycoordinating memory bandwidth sharing. At the storage level, we enableoperating systems to effectively exploit "sequential locality" and let random accesses to stay in buffer cache or/and in SSD (flash memory storage). We will mainly discuss the research issues and challenges in multicore processors in this talk. ================================Bio of the SpeakerXiaodong Zhang is the Robert M. Critchfield Professor in Engineering,and Chairman of the Department of Computer Science and Engineering atthe Ohio State University.His research interests cover a wide spectrum in the areas of highperformance and distributed systems. Several technical innovations andresearch results from his team have been adopted or being developed incommercial products and open source systems with direct impact to somekey computing operations, including the permutation memory interleavingtechnique first in the Sun MicroSystems' UltraSPARC IIIi processor andthen in the Sun's dual-core Gemini Processor, the token thrashingprotection mechanism and the Clock-Pro page replacement algorithm formemory management in the Linux Kernel and NetBSD. He has been elected as an IEEE Fellow for his "contributions to computer memory systems". Xiaodong Zhang was the Director of Advanced Computational ResearchProgram at the National Science Foundation, 2001-2004. He is theassociate Editor-in-Chief of IEEE Transactions on Parallel andDistributed Systems, and is also serving on the Editorial Boards ofIEEE Transactions on Computers, IEEE Micro, and Journal of Parallel andDistributed Computing. He is the program chair or co-chair in ICPP'07, WWW'08, and ICDCS'09. He received his Ph.D. in Computer Science from University of Coloradoat Boulder, and his B.S. in Electrical Engineering from Beijing University of Technology.=========================--※ 来源:·水木社区 http://newsmth.net·[FROM: 61.149.198.*]

没有评论: