PUMA
Istituto di Scienza e Tecnologie dell'Informazione     
Borgo R., Pascucci V. Distributed oriented massive data management: progressive algorithms and data structures. The document has been submitted to Journal : Data Science Journal, Codata, Technical report, 2003.
 
 
Abstract
(English)
Projects dealing with massive amounts of data need to carefully consider all aspects of data acquisition, storage, retrieval and navigation. The recent growth in size of large simulation datasets still surpasses the combined advances in hardware infrastructure and processing algorithms for scientific visualization. The cost of storing and visualizing such datasets is prohibitive, so that only one out of every hundred time-steps can be really stored and visualized. As a consequence interactive visualization of results is going to become increasingly difficult, especially as a daily routine from a desktop. High frequency of I/O operations starts dominating the overall running time. The visualization stage of the modeling-simulation-analysis activity, still the ideal effective way for scientists to gain qualitative understanding simulations results, becomes then the bottleneck of the entire process. In this panorama the efficiency of a visualization algorithm must be evaluated in the context of end-to-end systems instead of being optimized individually. There is a need at system level to design the visualization process as a pipeline of modules able to process data in stages creating a flow of data that need themselves to be optimized globally with respect to magnitude and location of available resources. To address these issues we propose an elegant and simple to implement framework for performing out-of-core visualization and view dependent refinement of large volume datasets. We adopt a method for view-dependent refinement that relies on longest edge-bisection strategies yet introducing a new method for extending the technique to the field of Volume Visualization while keeping untouched the simplicity of the technique itself. Results in this field are applicable in parallel and distributed computing ranging from cluster of PC's to more complex and expensive architectures. In our work we present a new progressive visualization algorithm where the input grid is traversed and organized in a hierarchical structure (from coarse level to fine level) and subsequent levels of detail are constructed and displayed to improve the output image. We uncouple the data extraction from its display: the hierarchy is built by one process that traverses the input 3D mesh while a second process performs the traversal and display. The scheme allows us to render at any given time partial results while the computation of the complete hierarchy makes progress. The regularity of the hierarchy allows the creation of a good data-partitioning scheme that allows us to balance processing time and data migration time still maintaining simplicity and memory/computing efficiency.
Subject Scientific Visualization
Out of Core Algorithms
76M27 Visualization algorithms


Icona documento 1) Download Document PDF


Icona documento Open access Icona documento Restricted Icona documento Private

 


Per ulteriori informazioni, contattare: Librarian http://puma.isti.cnr.it

Valid HTML 4.0 Transitional