Guía para la Implementación de análisis en paralelo de modelos dinámicos y estáticos No lineales mediante OpenSees

Abstract

Structural engineering research seeks to predict the behavior of structures, usually similar analyses are performed one after the other, called sequential analysis. However, it is more efficient to perform parallel analysis, so the computer performs several simultaneous processes, reducing both resolution times and energy costs. In spite of this, there are no documents that synthesize the most common methodologies through practical applications of this type of analysis. For this reason, a guide was developed for the implementation of parallel analysis for nonlinear dynamic and static models through OpenSees, using both personal computers and clusters. In this work, incremental dynamic (IDA) and static (Pushover) analyses were performed with a NIST prototype of the ATC-76-1 project, and the durations, speedups, and efficiencies of various methodologies were analyzed in order to identify the advantages and disadvantages of each one. Parallelization (Tcl) and multiprocessing (Python) are the simplest methodologies to implement, therefore the most recommended. However, OpenSeesMPI (Tcl) or mpi4py (Python) are useful to achieve greater process control. The most recommended cluster is CEDIA for its computing power, however, the management of other clusters such as Google Cloud or AWS is similar, so the guide can be used for the management of other servers to which the user has access. Furthermore, the commands used in the analyses, the installation and execution processes of each methodology were detailed, and video tutorials were developed for the implementation of each one of them, in this sense, it is expected that in future works these tools will be used to simplify the research work, and it is proposed to investigate more powerful tools than those proposed such as graphics cards for multiprocessing tasks with OpenSees