Simulating Cloth Free-Form Deformation with a Beowulf Cluster^*

Abstract

In this paper, we present the results of a parallel implementation of an algorithm which performs physical-based cloth free-form deformation, using a particle system modelling method. This sequential algorithm is characterised by being iterative over a large data set of coupled particles, which discretize cloth surface, and also of local nature in its mechanical simulation CPU-intensive method: the law of motion of each particle only depends in its local neighbourhood and is evaluated by solving an Ordinary Differential Equation (ODE). This characteristic makes the sequential algorithm suitable for a parallelisa-tion. We have made a parallel algorithm implementation of the mentioned problem exploiting the computing capacity of a Beowulf Cluster. We also propose a simple load balancing method for this algorithm. The results show that substantial speedup gains can be achieved by dividing the particle system domain into several processes. To objectively evaluate the results obtained with the parallelisation technique, we have utilised known metrics related to measure the overall performance of the parallel algorithm such as Relative Speedup.

This work is partially funded by the European Community IST Project Fashion-Me: “Fashion Shopping with Individualized Avatars”