Parallel Volume Rendering for Ocean Visualization in a Cluster of PCs

Volume rendering techniques can be very useful in geographical information systems to provide meaningful and visual information about the surface and the interior of 3D datasets. For ocean visualization, in particular, volume rendering techniques improve the analysis of the ocean inner structure, by generating visual information about, e.g., its temperature, salinity, velocity and mass. The rendering of huge datasets, however, is a computationally intensive task and, in order to achieve interactive visualization times, a high-performance computational system is fundamental. Although parallel machines have been successful in providing interactive times, most recent efforts have been directed towards a more cost-effective solution: implementing volume rendering algorithms on clusters of PCs. This platform has low-cost and can be easily upgraded. Parallel rendering applications, however, usually suffer from high load imbalance during the execution. In this paper, we propose a low-cost and high-performance system for ocean visualization in a cluster of PCs, DPZSweep. Our solution spreads the computation over the cluster and provides dynamic load balancing with a low overhead. Our experimental results show that when we included the load balancing algorithms, DPZSweep obtained up to 95% of parallel efficiency in 16 processors.