Abstract

In this paper, the new hybrid parallelizable low order algorithm developed by authors is coded in C and MPI, and is implemented numerically on parallel computing systems. To reduce communication overhead, different parallelisms and parallel solution strategies are used for forming and solving equations of motion and constraint equations. The presented implementation can accommodate the general spatial motion of a chain system allowing rotations and translations between neighboring bodies. Different numbers of key joint separations are used to study the influence of increasing dimension of system constraint equations on the overall computational efficiency of the algorithm. Numerical results for implementations of the spatial chain system show that the cutting of certain user-specified joints rather than every joints offers better computational performance. Comparisons between full sequential recursive implementation and parallel implementation of the same algorithm are presented in the form of speedup. Numerical results show that significant speedup can be obtained.

This content is only available via PDF.
You do not currently have access to this content.