A comparison of three different linear order multibody dynamics algorithms in limited parallel computing environments

Abstract

A comparison of three multibody dynamics algorithms, Featherstone's Divide and Conquer Algorithm (DCA), a fast sequential recursive algorithm, and an efficient version of the DCA (DCAe), in a parallel computing environment is presented. The comparison is made both in terms of theoretical (ideal) performance and with an implementation evaluated on a parallel computer. The DCAe was motivated by the promise of practical performance gains, and its development as a combination of DCA and efficient recursive methods is reviewed. An implementation of each method in C++ using an object oriented architecture for a shared memory parallel computer is described and results are reported for a computer with four processors. Benchmark results agree closely with the predicted algorithm performances. The default DCA is shown to require four processors to obtain results similar to the recursive sequential results and the DCAe is verified to outperform the efficient recursive solution with just two processors when load balancing is considered.