Usage of a computer cluster for physics simulations using bullet engine and OpenCL

Abstract

The study focuses on using a computer cluster for implementing real-time physical simulations, responding to a growing need for such use in various sectors, including medicine, video processing, automated transport management, robotics, and visualisation. The object of research is cluster and cloud technologies for conducting costly physical simulations for specific sectors, particularly high-budget and entertainment ones, such as cinematography and interactive entertainment. Research methods include using a modified Bullet engine to carry out physical simulations, integrated with OpenCL to work with the cluster. The choice of these technologies was determined by their high performance and adaptability to cluster systems. The research was based on a typical Bullet framework’s benchmark falling tower scene with the primary goal of measuring computational performance in frames per second. Results showed that the use of clusters is not advisable in environments with a low network throughput and the use of non-uniform computers. Under those conditions, simulations using a cluster become unstable with many objects and contacts between them and show a degradation in performance by an average of 50–60 % (to values of 10–20 frames per second). Despite the intermediate results of calculations on the cluster, the study met the expectations within the goals set and resources available. These results have significant implications for the further development of cluster and cloud technologies in physical simulations, providing valuable information about the limitations and capabilities of these systems.