VISUALIZATION: Efficient Out-of-Core Isosurface Extraction from Large Datasets

University Of Alabama In Huntsville, Huntsville AL

Investigators

Abstract

Isosurface extraction and rendering is a useful and popular method for exploring volume datasets. While many studies of extraction techniques have been presented, few researchers have considered how to perform isosurface extraction from very large datasets in a way that utilizes parallel computation while also effectively managing memory access. For out-of-core datasets (i.e., those too large to be processed entirely in main memory), delays from access of secondary storage must be minimized if high performance is to be achieved. In this project, an investigation of new techniques for parallel, out-of-core isosurface extraction are conducted. The work seeks to exploit multiple types of parallelism, effectively organize memory access to minimize access penalties, and to effectively manage inter-process communication. A hallmark of the workis that it focuses on total system performance rather than attempting to only maximize intermediary measures of a single aspect of performance. One primary target platform for testing of the techniques is cluster computation environments comprised of commodity CPUs. Intellectual Merit. The proposed activity can benefit multiple disciplines. Many scientific and engineering enterprises generate and/or wish to use large datasets. Some finance and consumer applications also desire to effectively utilize large collections of business data. Discovery of trends, phenomena, and structures in those datasets can be aided by more efficient visualization. In particular, for large datasets too large to be processed in-core, the time to compute a visualization is likely to be very high due to the relatively slow access times of the secondary storage. Reduction of these times by resource (memory, CPU, and communication)-effective processing will increase productivity among scientific visualization users and consumers. In addition, reduction of processing times may make tractable the consideration of certain very large problems. The proposed activity builds upon prior work of the PI and of the other experiences with parallel out-of-core isosurface extraction. The PI has access to the necessary computer resources to complete the work, including access to cluster computers at three sites and to supercomputers at three other sites. Due to the impact of the proposed activity on visualization methods that are used across the scientific community, society at large is likely to benefit via new knowledge discovery by the discipline science that this project aids. In addition, the proposed work will be beneficial in managing and understanding the increasing body of data being collected about scientific, engineering, and business phenomena. The project's results will be disseminated via publication in the open literature, in conferences, and on the web. Through the PI's partnerships with NASA, NIH, and other researchers, the results of the work have a high probability of producing an impact in multiple disciplines. The project will also aid undergraduate and graduate student training and development via (1) PI-mentoring of the graduate student supported on the project, (2) presentation, discussion, and analysis of results at a regular research forum, and (3) integration of research findings in graduate computer graphics/visualization courses taught by the PI.