arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1209.0960math.NAcs.DCcs.MScs.NA
收藏

基于聚集的大规模并行代数多重网格预条件子求解非齐系数椭圆问题

A Massively Parallel Algebraic Multigrid Preconditioner based on Aggregation for Elliptic Problems with Heterogeneous Coefficients

Markus Blatt, Olaf Ippisch, Peter Bastian

摘要:

提出了一种基于非光滑聚合的大规模并行代数多重网格方法。它特别适合于解决异构椭圆问题,因为它使用贪婪的启发式算法进行聚合,检测系数的变化,并防止聚合。在每个进程上使用解耦聚合,在粗略级别上将数据聚集到更少的进程上,它在解决方案的总时间和每次迭代的时间方面都可以很好地扩展到近300,000个核心。由于简单的分段常数插值之间的水平,它的内存消耗低,并允许解决超过100,000,000,000自由度的问题。

英文摘要:

This paper describes a massively parallel algebraic multigrid method based on non-smoothed aggregation. It is especially suited for solving heterogeneous elliptic problems as it uses a greedy heuristic algorithm for the aggregation that detects changes in the coefficients and prevents aggregation across them. Using decoupled aggregation on each process with data agglomeration onto fewer processes on the coarse level, it weakly scales well in terms of both total time to solution and time per iteration to nearly 300,000 cores. Because of simple piecewise constant interpolation between the levels, its memory consumption is low and allows solving problems with more than 100,000,000,000 degrees of freedom.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑