SHF:Medium:Collaborative Research:A comprehensive methodology to pursue reproducible accuracy in ensemble scientific simulations on multi- and many-core platforms

科学现象的集成模拟通常在高性能计算集群上运行数周甚至数月。这些计算环境的并发性已经很高，预计在不久的将来会显著增加，这不仅会使模拟遭受由于有限的算术精度而导致的数值误差，还会受到与多线程相关的执行中的不确定性的影响。最终，这种趋势会损害模拟结果，破坏科学界对集成模拟的信任。该项目解决了这个问题，并定义了一种方法，可以在包括多核和多核处理器的百亿亿级平台上实现大型集成模拟的可重复精度。该项目沿着两条主要战线推进。首先，研究人员确定准确性错误的常见来源，并在受控环境中研究它们的积累、传播和运行时影响。这一阶段包括三个研究活动：(i)生成代码基元，为可能导致准确性错误的计算建模；（ii）针对不同的并行平台，提供这些主题的多种实现，称为代码检查器；（iii）使用各种数据集和压力条件评估这些实现的准确性和运行时间。其次，通过在真实的科学代码库中安装这些代码检查器，研究人员可以研究它们在不确定环境中的行为。这一阶段包括两项研究活动：(i)根据定量影响分数对代码段进行优先排序，并将代码段与检查器主题匹配；（ii）找到最优的代码检查器实现，并用它们修补代码，以优化总体结果方差。本项目的目标应用是确定性混沌应用，包括n体原子系统模拟和天体物理模拟。

Ensemble simulations of scientific phenomena typically run for weeks or even months on high-performance computing clusters. The already high level of concurrency of these computing environments is expected to significantly increase in the near future, causing simulations to suffer not only from numerical errors due to limited arithmetic precision but also from the non-determinism in the execution associated with multithreading. Ultimately this trend can compromise the simulation results and break the scientific community's trust in ensemble simulations. This project tackles this problem and defines a methodology to enable the reproducible accuracy of large ensemble simulations on exascale platforms that include multi- and many-core processors. This project moves along two major fronts. First, the investigators identify common sources of accuracy errors and study their accumulation, propagation, and runtime effects in a controlled environment. This phase includes three research activities: (i) generating code motifs that model those computations that may lead to accuracy errors; (ii) providing multiple implementations of these motifs, called code inspectors, targeting different parallel platforms; and (iii) evaluating the accuracy and runtime of these implementations using a variety of datasets and stress conditions. Second, by installing these code inspectors in real scientific code bases, the investigators study their behavior in uncertain environments. This phase includes two research activities: (i) prioritizing code segments based on quantitative impact scores and matching segments to inspector motifs; and (ii) finding the optimal code inspector implementations and patching the code with them so as to optimize the overall result variance. The applications targeted in this project are deterministic chaotic applications including n-body atomic system simulations and astrophysical simulations.