Scalability Study of Explicit FInite Element Method on Shared and Distributed Memory Architectures

Parallel architectures in computer systems are powerful tools for processing enormous amounts of information. The results of this research will illustrate their optimal usage by evaluating the scalability of the code Abaqus in the cases of domain decomposition and loop parallelization architectures. The scalability was analyzed using concomitant variation having the DoF (Degrees of Freedom) and CPU (Central Processing Unit) numbers as the variables. Abaqus jobs with less than 100,000 DoF showed similar scalability than that of jobs with more than 500,000. These results contradict the original hypothesis which proposed that small jobs would show poor scalability and vice versa. Another evaluation will be realized with jobs containing more than 1,000,000 DoF to determine if scalability is actually related to the number of DoF