Designing Spacecraft High Performance Computing Architectures

Recent developments in space applications have indicated that future spacecraft computing platforms will demand for high performance computing (HPC) capabilities. In order to face this challenge, HPC technologies have to be introduced in the design process of such platforms. This paper summarizes some efforts taken to achieve this. A theoretical design for future spacecraft computing platforms is proposed. This design combines traditional reliability techniques and novel HPC solutions for efficient high performance. System components are not specified in terms of type and quantity, but only a logical representation of the system is given. Benchmarking results are obtained on different parallel computing systems to help platform designers in further system specifications. A real space application that reconstructs a synthetic aperture radar (SAR) image is used to benchmark shared memory, distributed memory, and heterogeneous CPU/GPU systems. It turns out that distributed memory systems are a necessity for performance improvements, whereas heterogeneous CPU/GPU systems offer much more efficiency in terms of performance per power consumption, size, and heat dissipation.