Application-level Benchmarking with Synthetic Aperture Radar

Space-based radar applications continue to receive increasing interest from the research community, and new technologies are emerging that will help to make the vision of real-time, on-board, high-volume data processing a reality for next-generation space platforms. Isolated kernel benchmarks may not accurately capture true system performance in the context of a full application, and so application-level benchmarking is needed to measure true system performance. This research presents a full Synthetic Aperture Radar (SAR) application benchmark, written in ANSI-C and using only the Gnu Science Library (GSL) and Message Passing Interface (MPI) libraries. The SAR application forms high-resolution images from raw radar data, and incorporates many computational kernels of interest to the HPEC community. We present performance results from two different parallelizations of the application, executed on a cluster of PowerPC processors connected via Gigabit Ethernet. In addition to presenting the performance results on our system, the benchmark source code and example data is made publicly available.