摘要
提出一种按照计算域分解的并行化方法来构建等几何分析的刚度矩阵和右侧向量.将计算域分解成为若干个不相交的子区域,然后为每个区域分配一个处理器,所有处理器并行进行子区域上面的计算,所有处理器完成子区域的计算以后,使用一个快速的归并算法完成线性系统的装配.实验表明,本文提出的方法在8核的机器上可以达到6.46的加速比,能够在4秒左右的时间计算680万个矩阵元素个数.使用Intel MKL稀疏求解器来求解线性系统,本文的等几何分析求解器能够在大约10秒的时间内求解52万的自由度,本文的方法比ISOGAT速度要快上万倍.
This paper proposes a parallel method based on the computational domain decomposition to construct the stiffness matrix and right hand side vector for IsoGeometric Analysis. The parallel method decomposes the computational domain into a set of disjoint sub- domains. Each subdomain is allocated to one processor. All processors perform the computation on its subdomain paraileUy. After all the computation is completed, a fast merge algorithm is executed to complete the assembling of the linear system. This method a- chieved the speedup of 6.46 on an 8-core CPU. With this method, 6.8 million elements can be computed in about 4 seconds. With Intel MKL Sparse Solver, a system with 520 thousand degrees of freedom can be solved in about 10 seconds. The method proposed in this paper is 10 thousand times faster than ISOGAT.
出处
《小型微型计算机系统》
CSCD
北大核心
2013年第6期1396-1399,共4页
Journal of Chinese Computer Systems
基金
国家"核高基"重大专项项目(2009ZX01028-002-003-005)资助
国家自然科学基金项目(60833004)资助
安徽省高等学校省级自然科学研究重点项目(KJ2012A008)资助
关键词
等几何分析
并行计算
计算域分解
IsoGeometric analysis
parallel computing
computational domain decomposition