A Scalability Study on Multicore Cluster Systems of an AFRL Radar Frequency Tomography Imaging Code Written in MATLAB(r) for Parallel Execution Using Star-P(r)

The Radar Signal Processing Technology Branch of the air force research laboratory (AFRL), Sensors Directorate (Kevin Magde, Principal Investigator) is investigating development of radar frequency (RF) tomography technology for sensing applications. In conjunction with the recent arrival of a remote test facility, the quantity of collected input test data will increase, and the computational requirements will significantly increase. As the teampsilas work involves frequent algorithmic modifications, the team focuses its research writing programs in high-level languages such as MATLAB^reg. Applications are currently running on the teampsilas own small SGI Altix^reg system and exploit parallelism via Interactive Supercomputing, Inc.psilas Star-P^reg product. Anticipating increased computational requirements, the team would like to use department of defense (DoD) high performance computing modernization programpsilas (HPCMP) major shared resource center (MSRC) systems for some of their work and seeks recommendations moving forward. The team is also interested in processing data at a field location when it is being collected in an effort to improve productivity. This study examines the scalability of example research RF tomography Star-P/MATLAB codes on the Arizona State University Fulton high performance computing Saguaro system, which is an Intel^reg multicore Xeon^reg cluster with an Infiniband^reg interconnect; this type of system can be found at DoD HPCMP MSRCs. The study provides direction toward shortening the development cycle time for RF tomography imaging experiments, in addition to providing guidance for future computing purchases.