Using High Performance Computers in Distributed Interactive VisualizationApplications at CEWES

摘要

Traditionally, when one considers how visualizations are created in the realm ofhigh performance computing (HPC), the most common model has two distinct phases. Phase one, in which the data is generated and where the role of the supercomputer is simply that of a data firehose spewing forth oceans of data. In phase two, this usually massive amount of data is then transferred to a graphics workstation where an appropriate visualization application is run to produce renderings that will allow for the exploration of the information. Often there are two major problems encountered in this model. First, the data must usually be limited to accommodate the smaller workstation memories (RAM and disk). Second, in many cases the time needed to compute the visual images is unworkably long, making interaction with the visualization impossible. In the model described above the supercomputer is not part of the visualization application at all. Recent technological developments in high speed networks and communication protocols have served to bring high performance computers out of the shadows and into the spotlight with the other components of visualization applications. Tapping the computational power of supercomputers for visualization algorithms and exploiting the high-speed network's ability to move large chunks of data rapidly to and from high performance graphics workstations enables the creation of an environment where distributive, interactive and collaborative visualization applications are possible.