In one view, high performance and large data visualization is largely an exercise in the efficient use of bandwidth. Terascale visualization problems involve movement of large amounts of data from spinning disk to software components that transform the data into visible pixels. End-to-end, problems of this magnitude require a staggering amount of bandwidth and processing power. When the data, software components used for visualization, and users are separated by a network, then efficient bandwidth management becomes an even more critical concern. This chapter describes a highly specialized visualization application, Visapult, which performs high performance remote and distributed visualization. Visaput is perhaps best known for its ability to effectively utilize the network, having been used to win the High Performance Network Bandwidth Challenge at the annual IEEE/ACM Conference on Supercomputing three years in a row (2000-2002).

We present a novel approach for latency-tolerant delivery of visualization and rendering results where client-side frame rate display performance is independent of source dataset size, image size, visualization technique or rendering complexity. Our approach delivers pre-rendered, multiresolution images to a remote user as they navigate through different viewpoints, visualization or rendering parameters. We employ demand-driven tiled, multiresolution image streaming and prefetching to efficiently utilize available bandwidth while providing the maximum resolution user can perceive from a given viewpoint. Since image data is the only input to our system, our approach is generally applicable to all visualization and graphics rendering applications capable of generating image files in an ordered fashion. In our implementation, a normal web server provides on-demand images to a remote custom client application, which uses client-pull to obtain and cache only those images required to fulfill the interaction needs. The main contributions of this work are: (1) an architecture for latency-tolerant, remote delivery of precomputed imagery suitable for use with any visualization or rendering application capable of producing images in an ordered fashion; (2) a performance study showing the impact of diverse network environments and different tunable system parameters on end-to-end system performance in terms of deliverable frames per second.