Issue No. 04 - October-December (2003 vol. 9)
Jeffrey Mahovsky , IEEE
Luigi Benedicenti , IEEE
<p><b>Abstract</b>—In this paper, we present a Java-based software architecture for real-time visualization that utilizes a cluster of conventional PCs to generate high-quality interactive graphics. Normally, a large multiprocessor computer would be needed for interactive visualization tasks requiring more processing power than a single PC can provide. By using clusters of PCs, enormous cost savings can be realized, and proprietary “high-end” hardware is no longer necessary for these tasks. Our architecture minimizes the amount of synchronization needed between PCs, resulting in excellent scalability. It provides a modular framework that can accommodate a wide variety of rendering algorithms and data formats, provided that the rendering algorithms can generate pixels individually and the data is duplicated on each PC. Demonstration modules that implement ray tracing, fractal rendering, and volume rendering algorithms were developed to evaluate the architecture. Results are encouraging—using 15 PCs connected to a standard 100 Megabit/s Ethernet network, the system can interactively render simple to moderately complex data sets at modest resolution. Excellent scalability is achieved; however, our tests were limited to a cluster of 15 PCs. Results also demonstrate that Java is a viable platform for real-time distributed visualization.</p>
Real-time rendering, distributed computing, java, volume rendering, ray-tracing, fractals.
J. Mahovsky and L. Benedicenti, "An Architecture for Java-Based Real-Time Distributed Visualization," in IEEE Transactions on Visualization & Computer Graphics, vol. 9, no. , pp. 570-579, 2003.