Abstract

Recent scientific and engineering advances increase the demands on tools for high performance interactive visual exploration of large-scale, multi-dimensional simulation and sensor-based datasets. For example, earthquake scientists can now study earthquake phenomena in detail via “first principle,” physics-based, large-scale simulations in a time-volumetric space. Interactive visualization benefits the iterative scientific process to extract information from data and help scientists adapt their methods. Single-system visualization software running on high-end commodity machines can no longer sustain interactive browsing of these large science data due to their limited I/O and processing capabilities. A distributed and incremental approach is needed, to allow selective filtering of the parts of the data that the scientist wishes to view. In this paper, we introduce a flexible and extensible Grid-based visualization framework for visual browsing of spatially and temporally large datasets in a Grid environment. Our framework leverages Grid infrastructure for access to shared scalable computation and data storage. In particular, we assume commodity machines for both user-interaction and bulk processing, exploiting distributed parallelism to scale the input data-handling capabilities. We describe the application of our visualization framework with data from the Southern California Earthquake Center ITR project, a large Grid-based science project. We report nearly ideal scaling results from a controlled experiment on small-scale, commodity, parallel storage and processing resources.

Date

January 1, 1970

Authors

M Thiebaux, H Tangmunarunkit, K Czajkowski, C Kesselman

Journal

Submitted to IEEE symposium on High Performance Distributed Computing