Product Code: ICAL09_P145

Accelerating the Simulation of Optical Distortion Due To Atmospheric Scintillation Using Multi-Core CPUs, GPUs, Cell Broadband Engine and FPGAs
Authors:
Vinay Sriram, Imperial College London; Adelaide Australia
Oskar Mencer, Imperial College London; - Australia
David Kearney, Reconfigurable Computing Laboratory, University of South Australia; - Australia
Presented at ICALEO 2009

Fast simulation of optical distortions due to atmospheric turbulence is important for simulating adaptive optics configurations, evaluating the performance of laser designators and simulating sensor target threat engagements. The current best published algorithm by Harding et. al. requires 600 seconds for a single simulation when executed as a single threaded on a Intel Xeon quad core processor. In this paper we present our attempts to first optimize this simulation in software and to then accelerate it using two Intel Xeon quad core 2 GHz CPUs, a Nvidia GeForce 9800 GX2 GPU, a 3.19GHz Cell Broadband engine, an array of Spartan 3E FPGAs and a heterogeneous computing cluster that comprises of two Intel Xeon quad core 2 GHz CPUs, one Nvidia Tesla C1060 GPU and V5LX330 FPGA board. Our optimized software implementation is 4 times faster. Our best multi-core implementation gives us a speed up of 7.6 times, while our GPU and Cell Broadband engine implementations give us a speed up of 5 and 4 times respectively over the optimized single threaded version. Our FPGA implementation delivers a 74 times speed up. Our best implementation on the heterogeneous computing cluster achieves an overall speed up of 300 times over the original single threaded implementation when the computation is distributed 300