文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Hierarchical algorithms and architectures for parallel scientific computing
侵权投诉

Hierarchical algorithms and architectures for parallel scientific computing

There has been a recent emergence of many interesting and highly efficient hierarchical (multilevel) algorithms (e.g. multigrid, domain decomposition, wavelets, multilevel preconditioning, the fast multipole algorithms, etc.) for solving numerical problems

Hierarchical Algorithms and Architectures for Parallel Scienti c Computing Tony F. Chan April 16, 1990

Abstract There has been a recent emergence of many interesting and highly e cient hierarchical (multilevel) algorithms (e.g. multigrid, domain decomposition, wavelets, multilevel preconditioning, the fast multipole algorithms, etc.) for solving numerical problems in scienti c computing. These algorithms derive their e ciency from using a hierarchical approach to capture the sharing of global information which is inherent in the physical processes being modelled. In addition to being computationally e cient, these algorithms also possess relatively high degrees of parallelism. I therefore argue that the architectures of parallel computers (especially massively parallel ones) should be designed to support hierarchical communication and synchronization needs of these algorithms. Dept. of Mathematics, Univ. of Calif. at Los Angeles, CA 90024. The author has been supported in part by the Army Research O ce under contract DAAL03-88-K-0085, the Department of Energy under contract DE-FG-03-87-ER-25037, and the National Science Foundation under contract NSF-DMS87-14612. Invited presentation to appear in the Proceedings of the 1990 ACM Int'l Conf. on Supercomputing, Amsterdam.

Hierarchical architectures are also more universal because they go beyond supporting a particular class of algorithms to supporting the underlying physical processes being modelled. Finally, I propose that algorithm designers take a critical look at traditional kernel algorithms and re-examine their coste ectiveness in a massively parallel computing environment.

1 Physics, Algorithms and Architectures There is a general agreement that parallelism is a cost-e ective way (some may argue it is the only way) to provide the increasing demand in performance in computational power in many areas of science and engineering. There is much less agreement, however, on how to design e ective parallel machines. It is well-known that it is not su cient to simply compare the peak computing rates of various designs. A successful architecture must strike a balance between the computational rate and the communication and synchronization overheads, the memory hierarchy and its bandwidths. Most importantly, the architecture must support e ective com-

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email