Cluster Computing:

MPT WINS DARPA CONTRACT FOR HIGH PRODUCTIVITY COMPUTING

Massively Parallel Technologies, Inc. (MPT) will compare its revolutionary Howard parallel processing technology to existing supercomputer architectures under a recently awarded Defense Advanced Research Program Agency (DARPA) Small Business Innovative Research (SBIR) program Phase II contract. MPT expects to achieve processing speedups of at least 450x with 511 nodes versus the 30x speedup objective of other DARPA High Productivity Computing System (HPCS) program participants. HPCS is focused on providing a new generation of economically viable high productivity computing systems and HPCS researchers have initiated a reassessment of how performance, programmability, portability, robustness, and productivity is defined and measured in the supercomputer realm.

MPT's Phase I study on high efficiency approaches for multi-sensor data fusion significantly exceeded objectives for processing efficiency while demonstrating operations on a parallel processing system scaled to ten times the size of the system defined for the study. MPT's original Phase II proposal addressed advanced data fusion techniques incorporating high resolution LIDAR imagery for automated feature extraction and change detection. At DARPA's suggestion, MPT changed its proposal focus to establish benchmarks for real world problems compatible with the Challenge Benchmarks developed for the HPCS competition between SUN, IBM, and Cray.

MPT's Phase I work, along with internal developments for image processing and bioinformatics operations, proves the validity of MPT's design. MPT can provide marketable high productivity computing solutions today while other HPCS solutions are not expected to be commercially viable until near the end of the decade. MPT's unique technology represents a major breakthrough in parallel processing performance. The company is able to integrate clusters of processors in a unique hardware topology that, along with MPT's parallelization software, accelerates processing speeds 250 to 500 times the speed of a single processor, instead of the 10 to 20 times speedup provided by today's conventional processing systems.