arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2110.09667math.NAcs.NA
收藏

低同步正交化方法在Anderson加速不动点求解器中的性能

Performance of Low Synchronization Orthogonalization Methods in Anderson Accelerated Fixed Point Solvers

Shelby Lockhart, David J. Gardner, Carol S. Woodward, Stephen Thomas, Luke N. Olson

AI总结:

针对Anderson加速(AA)因全局归约同步步骤多导致并行性能受限的问题,提出三种低同步正交化算法并集成到SUNDIALS的AA中,将每次迭代全局归约次数降至常数级,经CPU和多GPU实验验证,在并行强扩展极限下性能提升且收敛可靠。

摘要:

Anderson加速(AA)是一种用于加速非线性代数方程组不动点迭代收敛的方法。由于每次迭代都需要求解最小二乘问题,且依赖修正Gram-Schmidt方法更新迭代空间,AA需要额外的高成本全局归约同步步骤,且每次迭代的归约次数取决于迭代空间的大小。本研究将三种低同步正交化算法引入SUNDIALS(SUite of Nonlinear and DIfferential/ALgebraic equation Solvers,非线性与微分/代数方程求解器套件)框架下的AA中,将每次迭代的全局归约总次数降至2或3次的常数,与迭代空间大小无关。性能研究表明,在CPU大处理器数量场景下,新算法所需时间更短,且在多GPU架构上达到了预期性能。最重要的是,我们提供了多项数值实验的收敛性与时序数据,证明了这些算法在AA中的可靠性,以及在并行强扩展极限下的性能提升。

英文摘要:

Anderson Acceleration (AA) is a method to accelerate the convergence of fixed point iterations for nonlinear, algebraic systems of equations. Due to the requirement of solving a least squares problem at each iteration and a reliance on modified Gram-Schmidt for updating the iteration space, AA requires extra costly synchronization steps for global reductions. Moreover, the number of reductions in each iteration depends on the size of the iteration space. In this work, we introduce three low synchronization orthogonalization algorithms into AA within SUNDIALS that reduce the total number of global reductions per iteration to a constant of 2 or 3, independent of the size of the iteration space. A performance study demonstrates the reduced time required by the new algorithms at large processor counts with CPUs and demonstrates the predicted performance on multi-GPU architectures. Most importantly, we provide convergence and timing data for multiple numerical experiments to demonstrate reliability of the algorithms within AA and improved performance at parallel strong-scaling limits.

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。

↑