Collaborative visualization environment using P2P technology and ellipsoidal mesh partitioning

Abstract

A common technique to perform distributed (or parallel) rendering of a single frame is to break up a 3D scene and share the rendering load across multiple machines (called the rendering agents). The rendered sub-images from each machine are then composited on a single machine (called the compositor) and displayed on the screen (or saved to a file). The end result is an overall improvement in per frame render times for large and complex 3D models. However, this technique suffers from 2 major performance bottlenecks. Firstly, communication between the rendering agents and the compositor is heavy since depth information is also transferred with the rendered sub-images for every frame. Secondly, composition of all the sub-images from every rendering agent is an expensive process as every pixel has to be subjected to depth comparison. In this thesis, we propose a mesh partitioning algorithm (called Ellipsoidal Mesh Partition) and a mesh distribution algorithm (called Context Aware Mesh Partition) that eliminate the need for depth information for the compositing of the rendered sub- images. This reduces the compositing complexity. The key to both algorithms, is to break up a 3D mesh based on its unique features into smaller connected sub-meshes. If each sub-mesh is rendered by exactly one unique rendering agent, the composition of the rendered results will be equivalent to "piecing together a jigsaw puzzle". In other words, the compositing cost using our distributed rendering algorithm is reduced tremendously. Despite a minor (negligible) degradation in the final composited image, our results show an overall 40% performance improvement. Thus, we recommend this solution for use in distributed polygonal rendering applications and systems.