This Small Business Innovation Research Phase I project will determine the scientific, technical, and commercial merit and feasibility of developing a fast out-of core volume rendering software to visualize large multi dimensional data sets. Existing visualizers assume that the data to be rendered is RAM resident - consequently paging or explicit naive data movements between the RAM and the disk cause serious performance degradation. This constrains users to curtail the amount of volume data seen at a time and/or view the data at a coarser resolution. A novel combination of data structures, parallelism, compression and systems level optimizations will be used to render isosurfaces of large volume data sets at a very high speed. A simple proof of concept implementation, using optimized disk data structures and multithreading seems to corroborate the hypothesis that isosurfaces are rendered for large data sets at a speed that surpasses by orders of magnitudes the performance realizable from current visualizers when the data size exceeds the RAM capacity. This effort focuses on investigating further performance enhancements using data compression, parallel execution capabilities of the PC MMX-like instructions and additional optimizations of the data structures on the disk. It is anticipated that the product developed from this effort will dramatically improve the speed of viewing and visual navigation of large multi dimensional data sets. Consequently, it will have a significantly influence on medical imaging, drug synthesis, gene analysis and geosiesmic exploration capabilities.
