Time Optimization of Parallel Dynamic Analysis Using Greedy Algorithm in FEA

The Finite Element Method (FEM) is the most widely used numerical technique to predict the approximate response of a structure under various loading conditions. Predicting the response of a structure to seismic loading using FEM can be computationally intensive and time-consuming. Parallel FEM is one solution to such situations where the computation is distributed efficiently among multiple cores available in modern supercomputers. In order to utilise the advantage of parallel computing in FEM, Pacific Earthquake Engineering Research Centre (PEER), has developed the open source software, OpenSees, with advanced capabilities for performing parallel FEM specifically for carrying out earthquake engineering simulations. In this paper, a new methodology is proposed to improve the efficiency of parallel computation using greedy algorithm in OpenSees for the time history analysis of framed structures for multiple earthquakes. Greedy algorithm finds an optimal solution in a number of steps by effective scheduling and proper load balancing. This method is verified by studying the time required for analysis of arbitrary framed structures using a high performance computing machine with a 32-core CPU, 62-GB RAM and 256-GB memory. A percentage increase of 16.35 is observed in the speedup factor for a two dimensional model studied.