摘要
为提高分子动力学的模拟运算效率,在Linux环境下以MPICH技术构建的小型SMP集群系统上,对模拟体系采用改进的原子分解算法进行并行计算。将要模拟的原子平均分配给各个计算节点进行计算,节点间通过MPI进行通信。对进程间的接收和发送进行捆绑操作并采取非阻塞通信取代原有的阻塞通信,从而避免了死锁情况。实验结果表明:优化后的并行算法可以有效地利用计算机资源,提高运算效率,解决了实际测试中出现的死锁问题,在该集群系统上获得3倍以上的加速比。
To improve the efficiency of molecular dynamics simulation, a parallel computing cluster system was constructed with MPICH on Linux operating system, and the improved atom decomposition algorithm was adopted for parallel computing in the simulation system. The atoms to be simulated were distributed to to each node to calculate, and the nodes communicated each other via MPI. The processes to receive and send bundled, and non-blocking communication operations were carried out to replace the original blocking communication, thus avoiding the deadlock situation. The result indicates that parallel comuputing can effectively make use of computer resources, and improve the computing performance. Under the existing parallel systems, it would obtain parallel speedup more than three times.
出处
《计算机应用》
CSCD
北大核心
2010年第A01期235-236,240,共3页
journal of Computer Applications
基金
重庆邮电大学自然科学基金资助项目(A2008-39)
关键词
SMP
消息传递接口
并行计算
分子动力学
原子分解
Symmetric MultiProcessor （SMP）
Message Passing Interface （MPI）
parallel computation
Molecular Dynamics （MD）
atom decomposition
