基于Abaqus软件的并行计算集群平台构建与优化方法

根据有限元显式算法和隐式算法的特点，研究了岩土工程动力分析并行计算集群系统的硬件要求、集群系统的构建方法，构建了基于EM64T硬件构架、双路Intel Xeon处理器、Linux操作系统和64位Abaqus软件的32CPU并行计算集群平台，测试了存储子系统对集群性能的影响，比较了两种千兆以太网络作为、集群子网络的性能优化方法。以地下结构的地震反应分析为例，测试了优化前后该集群系统的计算速度，发现两种以太网络性能优化方法都可以有效提高集群计算速度。列举了该集群系统在深水桥梁基础流固耦合动力分析、地下结构地震反应分析和快速轨道交通环境振动分析中的应用，显示了该集群在显式算法、隐式算法及小规模、大规模数值计算问题中的并行计算效率，证明所构建的Abaqus数值模拟并行计算集群平台能够满足计算规模、计算精度和时效性的要求。

Construction and Optimization of Parallel Computing Cluster Platform Based on Abaqus Software

Based on the characteristic of FEA explicit algorithm and implicit algorithm, it is studied what parallel computing cluster platform＇s hardware is required and how the platform is built in the field of geotechnical engineering dynamical analysis, and the Abaqus software＇s numerical simulation parallel computing cluster platform is built, which is applied to dynamical analysis in the filed of geotechnical engineering. The cluster platform is based on EM64T hardware framework, uses dual-path Intel Xeon processor, makes use of Gigabit Ethernet subsystem, and configures 64-bit applications and Linux operating system. 32CPU parallel computing can be achieved using this cluster platform. The storage subsystem has impact on the clusterrs performance. The impact is tested. Two ways of Ethernet optimization are discussed. The platform＇s computing speeds before and after optimization are tested, using an underground structure dynamic analysis of earthquake FEA model. The test＇s result shows that the Ethernet performance can be optimized effectually, the two optimization＇s ways can increase the cluster platform＇s computing speed. The application examples in the field of fluid-solid coupling dynamical analysis, rapid rail transit ambient vibration analysis and underground structure seismic response analysis are introduced. The platform＇s computing speeds of explicit algorithm, implicit algorithm, small-scale calculation and large-scale calculation are showed through the application examples. It is proved that the numerical simulation cluster platform meets the requirements of calculation scale, calculation accuracy and timeliness.