High-performance computing has traditionally been the realm of lengthy, batch-oriented computations. As consumer technology has become more robust and feature-rich, clusters of commodity computers have emerged as a popular alternative to traditional monolithic supercomputers. Because these computers can easily be augmented with inexpensive consumer graphics processor units (GPUs), researchers have begun to investigate their use for high-performance interactive applications such as visualization. The computational power of a commodity GPU is increasing much faster than Moore's Law, and this, along with the GPU's inherent parallel vector-processing architecture, is making it an attractive platform for high performance general purpose computing. Unfortunately, clusters of graphics-enabled workstations have historically been underutilized because of their centralized nature; scientists must leave their normal work environment to access a dedicated "visualization center" containing a graphics-enabled cluster and exotic display and input devices. The investigator is solving this problem by enabling flexible remote access to these scalable, inexpensive clusters of commodity PCs. The ultimate vision is that a collection of centralized, easily maintainable and upgradeable, and inexpensive computers could serve the interactive, high-performance graphics and computation needs of an entire group, building, university, or even city. The research involves the design of appropriate visualization delivery infrastructure, as well as compelling multi-user, remotely accessible applications in the areas of engineering and education.
Leveraging his existing cluster-rendering infrastructure, the investigator is exploring two novel research directions. First, he is designing a remotely accessible graphics system that adapts to the capabilities of the target display. Using compressed video and streaming graphics API filters, he is creating technology to allow a single application to run effectively on displays with differing characteristics, such as resolution, color depth, brightness and form factor. This application need not be aware of the changing characteristics of the display; in fact a running application can be retargeted to a different display without restarting it. Secondly, he is creating the first ever network-aware graphics system, so that the rendering infrastructure and even the graphics architecture itself can adapt to the capabilities of the underlying network. For example, dynamic configuration of compression codecs will provide a consistent user experience as network bandwidth changes. To validate these ideas, the investigator will deploy a new visualization suite for mathematics education. Working closely with education researchers at the University of Virginia, he will design and field test new uses for visualization in the teaching of secondary mathematics. The availability of the remote delivery mechanism makes this an especially cost-effective way to bring judiciously applied high-performance 3D graphics into the classroom.
