 |
Inside this Issue |
|
| A Summer of Invention, Innovation and Immersion in High Performance Computing | ||
Dr. Frank Williams , ARSC |
||
Story by Lorien Nettleton
Each summer the staff of ARSC, along with participating faculty from UAF, welcome a group of undergraduate interns from diverse backgrounds to undertake research projects in the fields of computational science and high performance computing. The interns are teamed with UAF faculty mentors and charged with completing a 10-week project in one of a diverse array of disciplines, from sea-ice structure modeling, to engineering, to software and hardware design. For the interns, the experience is often their first time being immersed in a research environment, as well as being fully immersed in the wilderness environment of summertime in Alaska. Design of an Aerodynamic Lens Using Fluent
Stacey Schmidt, University of Wisconsin Madison Laser Raman spectroscopy is commonly used to determine the composition of an aerosol in the atmosphere. Dr. Javier Fochesatto has been interested in developing an aerodynamic lens device suitable for on-line aerosol sampling to assess the sample’s chemical speciation using this technique. Under laboratory conditions, an aerosol beam is created, and then irradiated by laser to measure the resultant Raman scattering of the solid and molecular species embedded in the aerosol beam. The aerodynamic lens device ensures a particle beam smaller than a laser beam, optimizing the irradiation cross section and allowing researchers to better analyze the composition of aerosol particles. Using Fluent, a computational fluid dynamics software package, Stacey Schmidt, of the University of Wisconsin Madison, created a simulation of a three-stage aerodynamic lens designed to emit a specific aerosol particle beam. The lens is a nozzle designed to hone a particle beam by pressure differences created by compressive flow stages within the lens. By staging three successive circular apertures, the air pressure is forced into incrementally focused airflow. The particle beam is formed as the inertia of the particles keeps them in the flow center line while the air molecules fall apart. At the outlet of the lens, an aerosol beam is created with dimensions and concentration suitable for particle spectroscopy. Schmidt also modeled a two-stage slit-geometry lens, which, in comparison to the three-stage circular aperture lens, resulted in a broader range of particle transmission and 10% higher transmission efficiency than the circular geometry for a particle diameter range of 0.2 to 3 microns. Improvement of Electric Potential Function in the UAF EPPIM Mark Wellons, the College of Wooster Solar winds are responsible for a number of space-weather phenomena, from aurora to magnetosphere generation. The UAF EPPIM relies on a number of inputs to generate space-weather simulations, and gathers those inputs from a variety of sources. Interns John Wright of Mercer University and Mark Wellons of the College of Wooster each worked on independent space weather modeling efforts to improve overall EPPIM performance. Parameters of solar winds, made up of ions, are essential for modeling space weather in the polar regions. Solar winds also make up a critical parameter in the EPPIM. Wellons undertook a project to simulate solar wind movement towards the Earth. The initial goal of his project was to analyze ion velocity and density data from NASA’s Advanced Composition Explorer (ACE) satellite, which collects information from solar winds as they pass through the LaGrange point where the magnetic fields between Sun and Earth are cancelled. These data are used to initialize a one-dimensional model that simulates the time delay for ions to reach the near-Earth WIND satellite. Once complete, the model is validated against actual travel time taken for ions to move from the ACE satellite to the WIND satellite. Information Retrieval Kylie McCormick, Mount Holyoke College As the volume of electronic documents continues to increase, the importance of obtaining relevant query results when searching for documents becomes essential. Kylie McCormick, of Mount Holyoke College, worked with Dr. Greg Newby to refine and evaluate methods for obtaining and merging relevant results from a distributed grid of interconnected sources into one optimized query result. As is commonly seen on Internet search engines, information retrieval systems retrieve documents in order of relevance using a variety of ranking schemes. But for each collection, those schemes may not be uniform. McCormick started with a naive merge algorithm that creates a new set of retrieved documents based on their aggregate significance. Using prior work by Newby and ARSC graduate student Chris Fallen, McCormick evaluated the adaptation of the algorithm’s search results. To evaluate multiple merge algorithms, McCormick used Simple Object Access Protocol (SOAP) to transfer information to and from index backends. She then used Apache Tomcat as middleware to store the multisearch servlet and Axis to design web services in Java to update the ARSC multisearch servlet interface. This new implementation will allow for the testing of different algorithms, as well as for analysis of different collections in a distributed grid information retrieval system. Finite Element Simulations of Compaction of Ice Particles
Matthew Poland, California State University Long Beach Matthew Poland of California State University worked with Dr. Jing Zhang to study the deformation of ice particles under compaction. This project is important to the understanding of the mechanical behaviors of sea-ice and road-ice, which impose threats to driver safety and road-structures. Poland simulated two-dimensional ice particles using ABAQUS/CAE, a commercial finite element package, running on the supercomputers at ARSC. After obtaining the mechanical properties for the density, flex and strength of ice, Poland arranged the ice particles in a series of grids constrained in a rigid container, where compaction was achieved through the stress of a rigid punch by uniaxial pressure. Two simulations, one with 36 particles and another with 98 particles were submitted to Iceflyer, one of ARSC’s IBM systems. Resultant stresses from the simulation were output and then visualized to display stress areas, with the intensity of the stress represented by color. Simulation results showed that the highest stresses occur at the contact points between particles; therefore, the initiation of fracture of the ice particles is expected at these points. Programming with Field Programmable Gate Arrays Keven Woo, California State University Long Beach ARSC is supporting researchers using Nelchina, ARSC’s Cray XD1 cluster, by testing software designed for the XD1’s field programmable gate arrays (FPGAs). Intern Keven Woo of California State Long Beach worked to evaluate the performance of FPGAs in comparison to the performance of a centra processing unit (CPU), using the Impulse C software development environment to simulate program executions on Nelchina. Woo also implemented several other programs on the FPGAs. As part of his work, he evaluated the Impulse C development environment’s performance in creating FPGA programs. He then investigated the design flow from this PC-based development tool, first testing and simulating the program on a PC, then actually running it on the XD1. Sea Ice Deformation in the Weddell Sea
Jennifer Hafer-Zdral, Reed College To track the constantly changing surface of pack ice in the Beaufort Sea, intern Jennifer Hafer-Zdral of Reed College worked with Dr. Jennifer Hutchings to analyze and visualize data collected from Global Positioning Satellite (GPS) buoys that are recording deformations of sea ice in the Arctic Ocean. Using Interactive Data Language (IDL), Hafer-Zdral read data sets generated by each GPS buoy into an array for processing and plotting. She then used the latitude and longitude for each buoy to provide a time-lapse image of the movement of leads, or fissures in the sea ice. To sort the data into time-lapse-ready segments, she reduced the data into a set of one-hour intervals and aligned the data collection by date. The IDL frames were turned into an animation by converting the output into Portable Pixel Map files and sequencing them to create an Autodesk animation. The animation was then combined with imagery collected from Synthetic Aperture Radar (SAR) to image the ice deformations for the period observed by the buoys. Each SAR image is stitched into the animation and overlaid onto the buoy location animation. Due to a failure of the georeferencing data from the SAR imagery, Hafer-Zdral had to resolve image misalignments by placing imagery into the animated mosaic manually. The result is an animation that makes it possible to see the pack-ice shearing and developing leads, while tracking the buoy movement data at the same time. Developing Sound in Virtual Reality Environments Sean Waite, Lycoming College ARSC has long invested in furthering the cutting edge of interactive virtual reality. This summer, intern Sean Waite of Lycoming College worked with UAF Music professor Dr. Scott Deal, graduate student David Krnavek and ARSC student employee Quinton Harris to explore the possibility of creating a virtual musical instrument that can be manipulated using body gestures in the environment of the Discovery Lab Virtual Reality Theater. Their project resulted in the Digital Audio Visual Environment (DAVE). A software musical instrument, DAVE is written in C++ and OpenGL and allows a user to play sound samples via physical gestures. DAVE uses HALCON camera imaging software to pinpoint a physical location within the environment. Cameras located in several locations within the physical environment view an object and merge the views to determine the exact location of the object. In this way, HALCON identifies the position of the input sensors–in this case LED lights wielded by the user in the Discovery Lab. By gesturing with one red LED and one green LED, the user is able to select a sound sample as an object and activate or deactivate a series of virtual buttons that control replay parameters such as volume, pitch and frequency of sound samples. Once a user selects a sound and assigns the replay parameters, a message is passed to Max/MSP, a music software package, which plays the sound sample along with the selected effects. Biofeedback Studies Using Virtual Reality Carlos Natividad, University of Texas at El Paso Dr. Boris Bracio engaged two interns to work on a project to create a simulation that could be used as therapy to help people overcome phobias. The project consisted of the development of two elements: a dynamic virtual environment in the ARSC Discovery Lab that would adjust based on an individual’s fear of the subject; and the hardware and network that will collect physiological data from an individual during the simulation and then modify the simulation based on that feedback. Dr. Bracio tasked the interns with developing a virtual environment that would increase or decrease the intensity of a stressful scene based on the subject’s pulse, blood pressure and temperature. The subject’s biometrics would be monitored, providing the data to control the simulation. Based on the subject’s response, the scene would become more or less intense—and therefore more or less frightening to those with targeted phobias—depending on the level of anxiety measured through the biometrics. This dynamic exposure, measurement and feedback process was dubbed the “fear factor.” The combination of hardware development and environment simulation makes this approach to psychological research unique, as most other research in this field relies on an operator controlling the fear stimulus. Carlos Natividad, of the University of Texas at El Paso, undertook the graphic design portion of the project, using the OpenGL language to program an environment that would stimulate a fear of heights in a subject. The simulation presents the user with an elevated path in a mountainous setting. As he or she walks through the setting, the path widens or narrows based on the anxiety level detected by the biofeedback hardware. To make the simulation realistic, Natividad worked to accurately map texture onto the trail and created a reverse fog that surrounds the user and fades as it extends into the distance, creating a snowy foreground effect. Kailah Davis, of the University of the Virgin Islands, developed the interface to collect the biometric data and to use the sensory information to create a formula that causes the virtual environment to change during therapy. Using a numerical matrix of possible configurations within the visualization and the intensity of the subject’s biometric data, the program forces the walls along the path to raise or lower and the path to widen or become more narrow, changing the perceived level of comfort for the subject. Improvement of Electric Potential Function in the UAF EPPIM John Wright, Mercer University John Wright worked to automate data acquisition and file archival for the UAF Eulerian Parallel Polar Ionosphere Mode (EPPIM). The EPPIM runs in two modes: real-time and post-analysis simulations. For real-time simulations, the model constantly updates the ionosphere prediction from currently gathered data. The real-time mode is able to produce predictions two hours before ionospheric events occur. For post-analysis simulations, filtered data is requested for a specific time frame and high-resolution results can be analyzed and compared to tomographic imaging data. Wright wrote scripts that search for and download the correct data for a given set of dates and parameters and then submit the data to the model for simulation. He also worked to enable quantitative comparisons between tomographic data and modeled data. Wright developed two new utilities, Datagrabber.py and Populator.py, and combined them with two previously existing utilities, Dataformer and Blade. Dataformer prepares model input data and calculates the post-analysis solar wind parameters; Blade captures the URL from output files used to compare tomography data. The new utilities developed by Wright collect data and store them in a single searchable database. Datagrabber.py, downloads three sets of raw data used to populate the model and Populator.py searches tomography imagery and data from the Northwest Research Associates (NWRA) database and stores the values in a local database. Finally, Wright automated the utilities to work together using the control script Oneyear.py. The Oneyear.py script performs five major steps that users previously had to do individually: obtain model input files from various servers; format all of the input for use in the model; write the configuration file; submit the job to the supercomputer queue; and monitor the model outputs, extracting the necessary data. |
||
| Challenges Index | ||
