Analysis of the Model Facility Geometry of the Visual Interactive Site Analysis Code Using the Java Programming Language. FREDRICK ROGERS (Norfolk State University Norfolk, VA 23504) RODNEY WRIGHT (Norfolk State University Norfolk, VA 23504) ROBERT SANDERS (Oak Ridge National Laboratory, Oak Ridge, TN, 37831) C. DAVID SULFREDGE (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Visual Interactive Site Analysis Code (VISAC) is a JAVA system that allows the user to simulate and analyze the results of different accidents/incidents ranging from simple individual equipment sabotage to complex sorties that utilize a range of military weapons, simulated truck or car bombs, or satchel charges. VISAC is supplied with a library of facility models that are customizable in both geometry and logic to approximate a number of facilities of interest. Created or customized model facilities can have faults, such as overlaps and voids in the geometry, which may be too small to be visible or even hidden by the complexity of a model. These faults may cause the production of erroneous results in the blast calculation portion of VISAC. The goal is to find such errors and report them so that faults can be easily corrected. To locate the errors, a grid of rays is projected through the model parallel to each of the three coordinate axes. Each ray is then divided into path segments representing the different material regions through which it passes. Interferences and voids are revealed by comparison of the path segment starting and stopping points at the region boundaries. This allows for correction of possible geometry faults.

Analyzing the Computational Complexity of Hierarchal Clustering Algorithms through the Use of Multiple Performance Metrics. PHILLIP MARTIN (Clemson University Clemson, SC 29634) THOMAS E. POTOK (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

With the advent of several methods for large scale text sorting and classification, it is essential to quantitatively compare each text clustering system. However, this comparison represents a significant challenge when considering the distributed nature and hierarchal structure of some of these methods. To overcome the difficulties in the analysis of each technique, it is necessary to examine the two most important aspects of a system, performance and quality. A system is impractical in the real world if it does not satisfy these two conditions. The IntelliAgents program for text clustering, developed by the Applied Software Engineering Research Group at Oak Ridge National Labs, was selected and then modified to include both a quality and a performance measuring algorithm. The quality measure is based on a modified version of the F-measure algorithm, which utilizes a calculation of precision and recall on document clusters to measure the quality of a clustering result. The performance of the system is observed by evaluating the time each document takes from its introduction into the system until the document is fully clustered. Selecting these methods provides several advantages, including ease of implementation and the applicability to a flat or hierarchal cluster system. Documents in each run consist of various subsets from the Reuters news 21578 text corpus. Testing is performed with 1 to 32 computer nodes in increments of 1 node. The data set is then varied from approximately 3000 to 6000 documents in increments of approximately 1000 documents. The results, as expected, show that as the number of nodes increases from 1 to 32, the time it takes to cluster x documents decreases. As nodes are added for each test, the F-measure of the system is relatively unchanged. The F-measure of the clusters only fluctuates with the change in the document set size. Having a simple and universal measurement, such as the ones in this paper, provides a basis for comparing clustering programs. With this foundation, it may be possible in the future to incorporate self-adjusting clustering algorithms that will balance performance versus quality to create optimally efficient clusters.

Assessment of Consumer Values - HVAC units that improve Indoor Air Quality. MATTHEW FIELDS (Trevecca Nazarene University Nashville, TN 37830) MELISSA VOSS LAPSA (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Heating, ventilating, and air conditioning (HVAC) systems are installed in buildings to regulate temperature and humidity. These systems provide air conditioning, controls for heating, and ducting, which ensures a uniform transfer of cold or hot air throughout a building. HVAC systems are the main energy consumers (more than 40%) in residential homes, according to Department of Energy's 2004 Buildings Energy Databook. According to a new study from The Freedonia Group, the need for HVAC systems is growing with annual growth rates of up to 2.4%. ORNL has an innovative idea for HVAC systems that will not only examine energy costs but will improve homeowners' health and comfort. The California Energy Commission (CEC) is interested in the impact of HVAC systems on homeowners' health. ORNL is working with the CEC to assess what is important to California residents about HVAC systems, including their perceived impact on indoor air quality in homes. Background research on residential HVAC systems and air filtration products is being performed to support this project. A price comparison of HVAC units and air filtration products available at home improvement centers was created to provide a baseline of data on available technology to improve indoor air quality. A database of intermediary and end-user contacts for the market assessment is being developed and discussions are being held with homeowners, utilities, HVAC contractors, home builders, and other organizations including the medical community. Data collected by the ORNL team will be compiled into a report for the CEC. The market assessment will provide the CEC with information needed to determine a path forward for sponsoring R&D on HVAC systems, based on potential consumer demand for advanced systems. This work builds from a successful ORNL project conducted last summer (Ashdown, B.G., Boudreau, G.D., Lapsa, M.V., Sherrod, M.J., Fields, M., St. John, C. and Wurster, D. "Assessing Consumer Values and the Supply-Chain Market for the Integrated Water Heater/Dehumidifier." ORNL/TM-2004/159, October 2004).

Developing Interactive Models of Industrial Facilities. JOSEPH LAKE (University of Tennessee Knoxville, TN 37996) ROBERT SANDERS (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

After September 11th, 2001, our nation has gained a heightened awareness of the need for greater national security. In order to protect our nation, we must not only protect the public, but also the facilities and infrastructure that permit normal day-to-day life for the population. These facilities include, but are not limited to: power plants, factories, military bases, and government buildings. Determining the best course of action for guarding these facilities and the security analysis of such structures is a daunting task. Due to the uniqueness of each facility, creating a general model is not always possible for real world scenarios. What is required is the ability to create separate models for each critical facility. These models need to be used many times over to generate a variety of different scenarios, and then the data from those scenarios needs to be assessed in a timely manner. Using the Unreal Tournament 2004 (UT2004) Engine, it is possible to make such models that will provide real-time data analysis for agent-based simulations. To create a representative model requires the software packages UnrealEd3.0 and Alias Wavefront's Maya 5.0 PLE, both included with the UT2004 distribution, and Discreet's 3D Studio Max 6.0. Employing these packages and detailed engineering drawings, a model of an industrial facility in the United States has been created. This model contains key systems (i.e. turbines, cooling stations, generators, etc) that will be linked to ORNL's VISAC fault tree analysis code to assess the result of damage to those systems from a variety of terrorist attack scenarios. These simulations will be run using the UT2004 Engine. Areas of interest in the simulations include, but are not limited to: possibility of meltdown, zones of reduced security, and the likelihood of disabling the plant. The ultimate objective of this methodology is to help identify and increase security in vulnerable areas, determine the vulnerability of key system components from internal and external attacks, visualize what subsystems are affected if an attack is successful, and estimate the probabilities of catastrophic events. In the future, manipulation of the Unreal Engine's code to allow for artificial intelligence capabilities that would mimic the behavior of known terrorist groups in an effort to create force-on-force simulations for training purposes may become possible..

Engineering Science and Technology Division Equipment and Calibration Database. RICHARD CLARK (Pellissippi State Technical Community College Knoxville, TN 37933) JOHN B. CZACHOWSKI (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

To increase the effectiveness in query and report generation and reduce undesirable redundancy and errors introduced in updating data, four separate flat-file databases, through three major revisions, were converted into a unified relational database. Work began with decomposition of the existing flat-file databases into their constituent parts. All the key attributes in the flat-files were identified, the repeating groups were removed and attributes were arranged to be dependent on each primary key. They were then further normalized to remove partial dependencies (attributes that were only dependent on one portion of a primary key) and transitive dependencies (which is a dependency of one nonprime attribute on another nonprime attribute). Microsoft Visio 2003 was the modeling tool used to design the structure of the relational database and the new system was built using Microsoft Access 2003. Lookup tables were identified and created to accommodate the one-to-one and one-to-many relationships present and transaction/bridge tables were created to further delineate the many-to-many relationships into needed one-to-many relationships. Once the basic tables were created the data from the existing flat-file system was extracted and scrubbed to remove inconsistencies and arranged in the proper order and format to be imported to the new relational system. Another aspect of the database was the inclusion of photos of the equipment to help identify items in a remote warehouse. This project will continue with the creation of queries that will be stored to speed up routine and repetitive searches commonly made of the system. From the existing tables and queries, data entry forms will be created that will allow for the management of the equipment, its placement in the laboratories and storage areas; pertinent, recurring calibration and routine inspection schedules of it by and for the equipment custodians.

Enhancing And Updating The Interactive Envelope Materials Database For Whole-Building Energy Simulation Programs. BRETT CARMICHAEL (Pellissippi State Technical Community College Knoxville, TN 37921) DR. JAN KOSNY (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Thermal resistance of a wall area is used to calculate the flow of heat from one wall surface to another. Decreasing the flow of heat (either containing the heat in cold weather conditions, or keeping it out in warm weather conditions), becomes the main focus of a more energy efficient building. Currently, an average of 45% of a building's energy expense is due to heating and cooling costs. Vast amounts of research have been done within this area, but accurate assessment of data generated has been a hindrance in the past. This is due to the widely varying composition of different buildings, including the details of the buildings, the different wall technologies used, and even the climate the building resides in. Due to the lack of commercial software which can adequately assess whole-building energy efficiency, creation of a user friendly online application has become the focus of the research. The development of the application focuses on three main areas. The first is the complex Whole Wall R-value Calculator. This calculator gives developers and builders the ability to easily discover the whole wall R-value for any house they are designing, simply by entering basic geometric dimensions and selecting from a list of tested wall systems. The most recent developments in this area include a simple Whole Wall R-value Calculator which runs over the internet and requires no downloading. Version 2 of the complex is also scheduled to be released. This newest version allows users to select multiple wall technologies within the same building along with implementing significant changes in the user interactivity levels. Single click wall selection and technology assignment allows the user to create complex building structures with little effort. The second area of focus is the Thermal Mass Calculator. It calculates the energy savings of a building based on the configuration of the wall technology and the climate in which the building is located. The final area is the online Hotbox Test R-value Database. This project contains years of hotbox data compiled at the Oak Ridge National Laboratories, and the goal of the database is to present the information in a clear, concise, and user-friendly way. The combination of these three areas helps to incorporate all the aspects needed to make an accurate assessment of whole-building energy efficiency.

Floating-Point Intensive Particle System Algorithms Implemented on the Graphics Processing Unit. JOSHUA STRATTON (University of Tennessee Knoxville, TN 37922) GEORGE FANN (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Graphical processing units (GPUs) have increased in programmability, speed, and computational capabilities. GPUs were originally designed for graphical purposes such as basic 2D rendering (GUIs, prompts, etc.) and 3D animations, while the central processing unit (CPU) controlled much more diverse tasks such as managing the operating system, peripherals and user applications. Recently GPUs have been designed specifically for physics-based games and have improved dramatically in floating-point capabilities-required in most 3D algorithms-allowing higher performance than most CPUs can provide. As the FLOPS/cost ratio of the GPU increases faster than the CPU, the GPU becomes a more and more practical tool for some types scientific computing. Using Cg, a particle-system algorithm written for the CPU has been ported to the GPU. This transition has improved the overall performance by taking advantage of the GPUs floating-point performance. Current GPUs can only perform 32-bit arithmetic, which is precise enough for many applications. Our paradigm is applicable to future 64-bit GPUs as well.

Forewarning an Epileptic Seizure through Parallel Computer Analysis. JASON COX (University of Tennessee Knoxville, TN 37996) KARA KRUSE (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Epileptic seizures can be forewarned by quantifying dynamical changes in nonlinear time-serial EEG data. The dynamical changes of an epileptic event are determined using a computer program to determine phase-space dissimilarity (PSD) measures. Computation of the PSD measures involves many different parameters which must be verified to find the set of parameters which gives the best forewarning predictions. Currently, the parameter optimization program can take several days to months to exhaustively test each parameter combination. Waiting this long to receive results makes progress difficult because only a slight change to the analysis or data means starting the program from the beginning. By using methods of optimization and high performance computing, the time to execute the program can be reduced considerably. Currently, an analysis is executed on a single data file multiple times. The file is read and stored into memory each time. Using large amounts of memory to store the file only once will result in the program running much faster than before. Since most of the data files are extremely large and multi-channeled, the use of a supercomputer can make it possible to store all of the data into memory and run multiple analyses on multiple processors. By reducing the time to run the program from months to days, progress towards obtaining the optimal parameters will be more feasible. Once the parameters have been determined, the program can run on a handheld computer and data can be obtained using a portable EEG device, giving the individual with epilepsy a chance to be forewarned of an impending seizure and potentially saving his or her life.

GWAVA: Java-based Information Retrieval on the Web for the Vitual Autopsy Project. JUDY EVANS (Roane State Community College Oak Ridge, TN 37830) DR. LINE POUCHARD (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Virtual Autopsy is part of the pre-autopsy process performed on combat fatalities. It is performed using a CT-scanner with three-dimensional capabilities. The initial images are sent to the Department of Radiologic Pathology to be further enhanced and interpreted. The final images are stored in a flat file repository. The additional physician findings are reported in the form of medical health records and entered into a database. Presently, retrieval of the database information and the corresponding images is done manually in several steps. The Generic Web Application for the Virtual Autopsy (GWAVA) is a system that follows the best software engineering practices of design, development and deployment. Medical doctors will be able to search certain fields of the database and retrieve relevant information from it as well as the corresponding images. For example a doctor will be able to search a particular description of a wound and the application would retrieve any information and images that pertain to that description. This information will then be used to research weapons and protection systems for soldiers. An architecture using Java Server Pages, Microsoft ACCESS and a Tomcat web server was designed and a prototype is being developed. Particular attention will be paid to secure access and user-based permissions. The medical community uses the Digital Imaging and Communications in Medicine (DICOM) standard. Therefore, the application requires programmatic access to a DICOM viewer. The application is being deployed on a localhost and the ORNL intranet and will be deployed over the web.

Investigation of Execution Environments for Cluster, Command, and Control (C3). WESLEY BLAND (Tennessee Technological University Cookeville, TN 38505) STEPHEN L. SCOTT (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

The Cluster Command and Control Tool Suite (C3) is designed to provide simplified functionality for users and administrators on a cluster of computers by providing a mechanism for operations on distributed machines in parallel. Included in this suite are tools to manage processes, transfer data, and provide support for other tools. Current work includes the implementation of the C3 tools as a client of the SciDAC Scalable Systems Software (SSS) project's Process Manager. The SSS environment uses a component-based systems software architecture for operating large-scale distributed computing systems. The integration of C3 into the SSS environment requires that C3 code be modified to communicate with the SSS Process Manager via the XML schema's specification over TCP/IP sockets. These changes to the C3 transport and execution model will enable the tools to take advantage of the scalability offered by the SSS components.

Parallel Computation of Blood Flow in Arteries with Moving Walls. HANNAH FLATH (Columbia University New York, NY 10027) RICHARD WARD (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

A computational model of realistic blood flow in a section of artery could improve diagnosis and treatment of health problems such as aneurysms and wounds to artery vessel walls. Systems of equations describing fluid and wall behavior are numerically simulated in PICMSS (Parallel Interoperable Computational Mechanics System Simulator), a parallel computational software environment for solving equations using finite element analysis. Currently under investigation are systems of equations published by A. Quarteroni et al. in Comput Visual Sci in 2000. These systems include Navier-Stokes and incompressible fluid equations coupled through boundary conditions to wall motion. PICMSS was modified to include these fluid-structure interactions. The algorithm presented by Quarteroni et al. was then implemented. The modifications to PICMSS were evaluated through comparison with results from an example of a 2D compliant pipe in the Quarteroni et al. publication. Further work is required to complete the modifications.

Reconfigurable Computing with the SRC-6: System Architecture, Performance Data, and Efficient System Utilization. DAVID WORSHAM (Georgia Institute of Technology Atlanta, GA 30332) JEFFERY VETTER (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

The SRC-6 is SRC computing's flagship hybrid computing system. Using a combination of DLDs (Dense Logic Devices), such as microprocessors, and DEL (Direct Execution Logic), such as FPGAs, this system can provide large performance gains over traditional computing systems for highly repetitive or computationally intensive problems. Performance data was collected on this system, including bandwidths for the various memory interconnects and timings for various aspects of executing code on the FPGAs, in an attempt to better understand how to utilize the system in an efficient manner. This data was acquired by performing work on the system, such as sending data across memory buses, and using a nanosecond accurate timer to measure how long this work took. This data confirmed SRC's numbers, which was expected. These results will be used as reference data in further papers concerning the SRC-6 system.

SensorNet: Combining the Existing to Protect the Future. ISAIAH RUFFNER (University of the Cumberlands Williamsburg, KY 40769) DAVID HILL (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

For as long as the United States exists, there will always be a need to manage the ever increasing threat of terrorism. That truth was not fully realized until the attacks on the World Trade Center and the Pentagon in September 2001. Unfortunately, it was only then that Americans realized the real need of early detection. With the rush to appease a demanding public, many agencies, cities, etc. implemented such systems without a true vision for their possibilities, thus resulting in many sensor networks that do little sharing of data (if any) with other networks. This is a problem that the SensorNet project seeks to resolve. The idea of SensorNet is to take existing networks of sensors, create additional networks where needed, and combine everything into one network that is able to access, store, and retrieve that information from any network enabled location. One immediate need for the SensorNet project was the lack of an application which would allow users to view events that had already taken place at transfer truck weigh stations. An application had already been developed by David Feaker that allowed an operator to look at the data in real time, however, it was not designed to allow reach-back capabilities. The SensorNet Weigh Station Viewer needed to be similar in design to the existing application to promote familiarity, and it needed to provide some additional functionality. The application was designed using the Java programming language and JavaWebStart technology, which allows a Java application to be installed and run remotely. It provides users with the ability to study the data taken, including data from various radiation detection devices, as well as provides images and data that provide physical descriptions of the trucks.

Using Logistical Networking Methodologies to Improve Long Distance Transfer Performance. DAVID LABISSONIERE (East Tennessee State University Johnson City, TN 37614) JEFFREY VETTER (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

It is often necessary to move large data sets over great distances across a Wide Area Network (WAN). Due to the design of the congestion control mechanisms in the Internet Protocol suite, the effective bandwidth obtained is often suboptimal. However, by employing Logistical Networking techniques, this performance can be improved. Logistical Networking is a paradigm which uses intermediate time-limited storage to provide buffering areas for data transfers. The foundation of Logistical Networking is the Internet Backplane Protocol (IBP) which allows allocation and access to temporary storage on remote servers. These servers are called depots and are located at strategic points around the Internet. IBP can be used to transfer data indirectly through one or more intermediate depots. To evaluate potential benefits of this technique, the researchers have designed and executed transfer tests of varying data sizes between two systems, both by passing data directly through the network and by passing data in pipelined fashion through an intermediary depot located along the direct network path. When comparing the results of these two transfer techniques, performance improvements can be observed when using the indirect approach when the data size is larger than approximately four megabytes. Even though additional processing is required at each intermediary depot, these costs are outweighed by the performance benefits garnered from shorter transfer distances. Although other potential optimizations have been identified, this work shows the capability of Logistical Networking to improve long-distance data transfer performance.