Parallelization of the CVODE ordinary differential equation solver with applications to rolling bearing simulation

Abstract

We discuss how to solve ordinary differential equations on parallel computers, with application to dynamic rolling bearing simulation. We show how to parallelize both the solver and the model, in order to get a scalable application. The obtained results show that, within the original CVODE solver, LU factorization and the forward/backward elimination of the Newton matrix, for the rolling bearing application can be done in almost constant time, independent of the problem size.