Tu Yi-Cheng, Chen Shaoping, Pandit Sagar, Kumar Anand, Grupcev Vladimir
Department of Computer Science and Engineering, University of South Florida, Tampa, Florida - 33620
Department of Mathematics, Wuhan University of Technology, 122 Luosi Road, Wuhan, Hubei, 430070, P. R. China
ACM BCB. 2012 Oct;2012:527-529. doi: 10.1145/2382936.2383010.
Analysis of large particle or molecular simulation data is integral part of the basic-science research community. It often involves computing functions such as point-to-point interactions of particles. Spatial distance histogram (SDH) is one such vital computation in scientific discovery. SDH is frequently used to compute Radial Distribution Function (RDF), and it takes quadratic time to compute using naive approach. Naive SDH computation is even more expensive as it is computed continuously over certain period of time to analyze simulation systems. Tree-based SDH computation is a popular approach. In this paper we look at different tree-based SDH computation techniques and briefly discuss about their performance. We present different strategies to improve the performance of these techniques. Specifically, we study the (DM) based SDH computation techniques. A DM is essentially a grid dividing simulated space into cells (3D cubes) of equal size (volume), which can be easily implemented by augmenting a Quad-tree (or Oct-tree) index. DMs are used in various configurations to compute SDH continuously over snapshots of the simulation system. The performance improvements using some of these configurations is presented in this paper. We also present the effect of utilizing computation power of Graphics Processing Units (GPUs) in computing SDH.
对大粒子或分子模拟数据的分析是基础科学研究领域不可或缺的一部分。它通常涉及计算诸如粒子间点对点相互作用等函数。空间距离直方图（SDH）就是科学发现中一项至关重要的计算。SDH经常用于计算径向分布函数（RDF），采用朴素方法计算需要二次时间。朴素的SDH计算成本更高，因为在分析模拟系统的特定时间段内需要持续进行计算。基于树的SDH计算是一种流行的方法。在本文中，我们研究不同的基于树的SDH计算技术，并简要讨论它们的性能。我们提出不同的策略来提高这些技术的性能。具体而言，我们研究基于数据网格（DM）的SDH计算技术。数据网格本质上是一个将模拟空间划分为大小（体积）相等的单元格（三维立方体）的网格，通过扩充四叉树（或八叉树）索引可以轻松实现。数据网格用于各种配置中，以便在模拟系统的快照上持续计算SDH。本文展示了使用其中一些配置所带来的性能提升。我们还展示了在计算SDH时利用图形处理单元（GPU）计算能力的效果。
