Using Clusters for Large Data Volume Rendering
Current graphics cards have good capabilities to display volumetric data in an interactive fashion, but only as long as they can keep the data in their local memory. With large data sets this is not possible. Our approach to these problems is using not one, but a cluster of computers, each looking at only a part of the volume.
Volumetric data is becoming more and more prevalent. CT and MRI scans are very common diagnostic methods, and are starting to be ubiquitous. They are also increasingly being used for non-medical applications like non-destructive material evaluation. As part of the technological development the amounts of data that they generate is growing steadily. In addition to these new areas there are classic areas with very large volumetric data problems, like oil and gas companies and large-scale simulations in meteorology and other scientific fields. At the same time clusters of simple personal computers are becoming an important force to drive tiled displays like the HEyeWall. Our goal is to use the combined memory and rendering power of these clustered systems to split up the large volume data and combine the partial volume's images to a single, coherent result. One problem in doing that is the required network bandwidth. Standard networks like GBit Ethernet become a bottleneck very quickly. An additional problem is getting the partial images out of the graphics cards. They are not designed for this, and most current cards have very limited bandwidth back into the main system. Both of these problem can be solved with a dedicated, video-based combiner. VRAC is in the lucky position of having dedicated video combiner systems, and we're working on using them to do the distributed rendering and partial image combining.
- Started:
- 2005-08-15
- Ended:
- 2006-05-15
- Past Students:
- Andy Menz
- Funded By:
- Iowa State University Research Grant ($40,000)
