Parallel Performance of Pure MATLAB "M-files" versus "C-code" as Applied to Formation of Wide-Bandwidth and Wide-Beamwidth SAR Imagery
Abstract
To the extent that tools can allow rapid prototyping, increased value is achieved. For scientists in the area of Signal and Image processing, MATLAB is often the tool of choice. MATLAB allows extremely rapid prototyping and debugging of complicated studies with a minimal of computer science expertise. Frequently an idea is implemented in MATLAB code and tested on small data sets. These small data sets provide outputs in sufficient time such that performance studies can be conducted. However, as the studies progress and the same codes are used on much larger real data sets, run times may grow to unrealistic lengths. This may force the research scientist to use another implementation language, such as C and MPI, and implement the code on parallel architectures. This code conversion is undesirable, time consuming, and error prone. An alternative and convenient way to accomplish run time reduction is to use the MatlabMPI suite written by Jeremy Kepner of MIT. This suite of pure MATLAB code allows one to simulate many of the MPI functions within MATLAB. It accomplishes inter-processor communication via disk I/O and a common disk drive.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 25, 2003
- Accession Number
- ADA428838
Entities
People
- Ashok Krishnamurthy
- John Nehrbass
- Juan C. Chaves
- Mehrdad Soumekh
- Stan Ahalt
Organizations
- Ohio State University