Enhancing the Space Fence Performance Evaluation System (PES):
GPGPU Computing in a Complex, Multithreaded Software System
HPEC 2010, MIT Lincoln Laboratory, Lexington, MA, September 15, 2010
Satellite simulation is a computation intensive component of the Space Fence Performance Evaluation System (PES). To improve the performance of the PES, we have integrated and enhanced a GPU-based implementation of SGP4 developed at MIT Lincoln Laboratory. Successful integration of GPU-based calculations into the PES, a complex system built using a Java–based, real-time streaming server infrastructure, required solution of several challenges that are not commonly discussed in the GPGPU computing literature. In addition to the traditional issue of optimizing memory access patterns , we solved the key problems of multithreaded access to shared GPU resources and integration of CUDA with a Java software framework.