Distributed Processing for Rapid ReconstructiOn of Terrain Models
Abstract
The project goals are to develop high performance computing techniques for rapid mapping of terrain and cultural sites to support a diverse set of capabilities for battlefield awareness. With the recent increase in the variety and resolution of image sources, there is an increasing need to rapidly generate specialized map products. To meet these requirements, the University of Massachusetts is developing parallel and distributed processing techniques. The first year research effort focused on mechanisms and strategies for using distributed processing techniques across a heterogeneous set of computer architectures, and parallel processing techniques for processing very large data sets. Several studies were conducted using the PVM (parallel virtual machine) library support. We successfully applied the technique to a complex perceptual grouping algorithm used in site model reconstruction. In addition, we began work on developing out-of-core algorithms that advantage multiprocessor servers to efficiently process very large data sets (images greater than 30,000 x 30,000 pixels). These algorithms circumvent the highly inefficient operating system virtual memory and paging algorithms. initial bench match studies have shown that the efficiency of these out-of-core techniques is independent of the image size.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 06, 1998
- Accession Number
- ADA357877
Entities
People
- Edward M. Riseman
- Howard J. Schultz
Organizations
- University of Massachusetts Amherst