A Joint STARS perspective on real-time high performance computer benchmarking

Abstract

A real-world application perspective on embedded real-time high performance computing benchmarks. Joint STARS (Joint Surveillance Target Attack Radar System) is an air-to-ground surveillance system which provides on-line real-time information on the detection, location, classification, and tracking of ground based objects over a large area. The real-time airborne nature of Joint STARS places physical as well as performance requirements on any computing system on the Joint STARS aircraft. Real-time high performance benchmarks must provide performance and utilization metrics in the context of real world system constraints. Programs such as Joint STARS could benefit from high performance computing benchmarks which: a) provide kernels that are representative of common fundamental application building blocks; b) use open specification and reporting guidelines such as PARKBENCH; c) have publicly available downloadable results and information on the World Wide Web; d) report the size, weight, and power required for a particular system to execute a given benchmark; e) give visibility into sustainable transfer rate characteristics of I/O interfaces; and f) include instrumentation for system utilization such as sustained to peak performance ratio, interprocessor communication bandwidth use, and memory consumption.