Interactive rendering of massive terrains on PC clusters

Abstract

We describe a parallel framework for interactive smooth rendering of massive terrains. We define a parallelization scheme for level of detail algorithms in cluster-based environments. The scheme relies on modern PC clusters capabilities to address the scalability issue of level of detail algorithms. To achieve this, we propose an eficient tile-based data partitioning method that allows both reducing load imbalance and solving the well-known border problem. At runtime level of detail computations are performed in parallel on cluster nodes. A hierarchical view frustum culling combined to a compression mechanism harnessing the frame-to-frame coherence are used to drastically reduce the inter-tasks communication overhead. We take into account level of detail algorithms visual quality issue by providing geomorphing and texturing supports. We are able to interactively and smoothly render terrains composed of hundreds of millions to billions of polygons on a cluster of 8 PCs.