Andrew Lindstrom is a senior studying Computer¬†Science at University of Hawaii at Hilo, but his¬†home is on the island of O’ahu. He enjoys¬†troubleshooting problems and working with his¬†hands. He plans to graduate with a Bachelor of¬†Science in Computer Science and to work as a¬†system administrator. Andrew likes playing games¬†and guitar during his spare time.

Home Island:¬†O‚Äėahu
Institute when accepted: University of Hawaii at Hilo

The Faster Tracker: Simulating Satellite Tracking Using Parallel Computing Project Site: Maui High Performance Computing Center Mentor: Carl Holmberg & Paul Schumache

Project Abstract: The task of tracking and detecting satellites orbiting the Earth is performed using a number of¬†ground-based sensors around the world, each one tasked as an isolated system. Objectives¬†may include tracking specific, known satellites, or detecting previously unknown objects. Both¬†types of observations are used to update a central catalog of satellites. This data collection¬†and catalog updating process is largely manual, and it may not scale sufficiently to handle the¬†growing number of objects to be tracked. However, networking the sensors together and tasking¬†them from a single facility may obtain the required scalability. This concept can be tested¬†using software models of the existing sensors. This project’s first step is to use an existing¬†MATLAB model in its alpha stage of development, and from it create a version that will take on¬†a larger data set without sacrificing turnaround time. The existing model simulates the FPS-85¬†radar. The model was analyzed to determine which types of performance optimizations would¬†likely provide the desired throughput using the computational resources at MHPCC, within the¬†time constraints of a summer internship. A data parallel approach was selected. This process¬†included taking advantage of the MATLAB Parallel Toolbox, removing user interactive features,¬†and adding output file I/O. Initial development was completed on a multi-core Windows-based PC using a 10-satellite test set before moving the work to MHPCC’s Linux-based¬†“Mana” cluster. The next task is to test the parallelized model’s scalability on Mana using increasing¬†numbers of CPUs and nodes. Final outcomes will include process timing, measurements¬†of performance vs. the number of processors used, and suggestions for more suitable¬†parallel-processing approaches using MATLAB Parallel Toolbox, if any. Future goals of this¬†project will include testing against a larger data set, and to test multiple models linked over a¬†network to a resource manager. Also, the original PC-based radar simulator would benefit¬†from additional optimization of the user interface and data displays.