arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1402.7216cs.DCcs.CEphysics.comp-ph
收藏

高度平行双金属结构的大规模分子动力学模拟

Large-Scale Molecular Dynamics Simulations for Highly Parallel Infrastructures

Jana Pazúriková

摘要:

计算化学允许研究人员在sillico中进行实验:通过对感兴趣的生物或化学过程进行计算机模拟。分子动力学采用分子力学相互作用模型模拟原子的N体问题$-$,根据牛顿物理和原子静电相互作用的经验描述计算原子的运动。这些模拟需要高性能的计算资源,因为每个步骤中的评估在计算上要求很高,并且需要数十亿个步骤才能达到有趣的时间尺度。目前的方法分解的空间域的问题,并计算并行/分布式基础设施。即使是具有最强扩展性的方法也达到了50万个核心的极限:如果提供更多的处理器,它们就无法缩短产生结果的时间。在大规模并行计算资源的exascale计算的黎明,我们希望通过将并行计算的分子动力学模拟,以提高并行的水平。在没有先验知识的情况下同时计算几个连续时间点的结果,没有取得重大成功。我们将研究和实现一种新的方法组合,根据我们的理论分析,与时间顺序计算相比,应该实现有希望的加速。

英文摘要:

Computational chemistry allows researchers to experiment in sillico: by running a computer simulations of a biological or chemical processes of interest. Molecular dynamics with molecular mechanics model of interactions simulates N-body problem of atoms$-$it computes movements of atoms according to Newtonian physics and empirical descriptions of atomic electrostatic interactions. These simulations require high performance computing resources, as evaluations within each step are computationally demanding and billions of steps are needed to reach interesting timescales. Current methods decompose the spatial domain of the problem and calculate on parallel/distributed infrastructures. Even the methods with the highest strong scaling hit the limit at half a million cores: they are not able to cut the time to result if provided with more processors. At the dawn of exascale computing with massively parallel computational resources, we want to increase the level of parallelism by incorporating parallel-in-time computation to molecular dynamics simulations. Calculation of results in several successive time points simultaneously without a priori knowledge has been examined with no major success. We will study and implement a novel combinations of methods that according to our theoretical analyses should achieve promising speed-up compared to sequential-in-time calculation.

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。

↑