生物大分子的分子动力学模拟过程在百万亿次集群上的部署优化

分子动力学模拟作为研究生物大分子功能和性质的新工具,已广泛应用于蛋白质和核酸等物质的分子动态学行为研究,但目前常规分子动力学模拟时间尺度较小,不能达到生物大分子分子动态行为的有效取样范围。温度副本交换分子动力学可同时运行多个独立模拟,明显提高模拟时间的可及尺度,但需要千核以至万核的计算资源,目前已发表的相关文献其模拟体系使用的计算资源均较小。本文利用国家超算济南中心的神威4000A百万亿次集群,首先进行单副本的分子动力学模拟,然后利用外切纤维素酶催化结构域的模拟体系(约5万原子)进行多达128个温度副本的分子动力学模拟,一次模拟任务最多成功利用6 720个CPU核心同时进行计算,最高总运算速度累计达到2 274 ns/d,这为分子动力学模拟利用上万核心进行计算提供了新的思路。

Deployment and optimization of biomacromolecule molecular dynamics simulation process on hundred trillion times cluster

As a novel tool to study the functions and properties of biomacromolecules,molecular dynamics has been extensively used to investigate the molecular dynamic behavior of proteins and nucleic acids.However,the time scale range of normal molecular dynamics simulation remains relatively narrow,which cannot reach the valid sampling range of biomacromolecules dynamics behavior recently.Temperature replica exchange molecular dynamics can run multiple independent simulations synchronously and effectively increase the simulating speed.Meanwhile,it requires tens of thousands of cores in aspect of the computing resources,yet the simulation systems of published literatures cannot reach such large scale hitherto.By using the 100T Flops cluster of sunway 4000A in National Supercomputing Center in Jinan,we firstly ran the molecular dynamics of single replica and then ran as many as 128 temperature replicas exchange molecular dynamics to simulate the catalytic domain of exocellulases(about 50 000 atoms).We simultaneously used up to 6 720 CPU cores in this task,the total computing speed accumulated to 2 274 ns/d,which offers a new approach to run the molecular dynamics over ten thousand CPU cores.