Student Abstracts: Computer Science

A Case Study of Efficient Social Network Simulation through General Processing on Graphics Processing Units. BRANDON AABY (Maryville College, Maryville, TN, 37777) KALYAN S. PERUMALLA PH.D (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Agent based simulation has been both a large area of study and a widely used tool for scientific research in past years. Current implementations run on standard CPUs, and with the requirement of processing ever growing data sets, higher computational speed is of the utmost importance. General processing on graphics processing units (GPGPU) is an emerging platform offering the possibility of increased speed for data sets and models that can be processed in parallel. Agent based simulation is one such candidate for performance gains in a GPGPU implementation. My research has focused on thoroughly investigating GPGPU’s suitability for providing researchers with a more efficient way of conducting agent based simulation research. Studies were done using two conventional models: two-dimensional diffusion and Conway’s Game of Life. I first created an optimized CPU diffusion model and, following a determination of accuracy, compared computational speed with an Open Graphics Library GPGPU implementation previously developed at Oak Ridge National Laboratory. Similarly, like studies were completed with the Game of Life. Following this strict CPU and GPGPU comparison, further comparisons and analyses were conducted with a widely used agent based simulation API, Repast. Evaluations involving Repast revolved around the premise that were GPGPU to be harnessed by researchers for agent based simulation, it must be competitive with currently used research technologies. Results obtained with both two-dimensional diffusion and the Game of Life show significant performance gains through GPGPU. For a plethora of data sizes, it has been found that the GPU processes the models in parallel at much greater rates than both optimized CPU code and Repast. Furthermore, as both sample size and the number of iterations through the model increase, the gap between GPU and CPU performance becomes even wider. These successful studies are to now be extended by investigating new models through GPGPU and exploring compatibility with necessary agent based simulation functions, such as image generation. Nonetheless, this research has certainly shown that GPGPU has the potential to become an efficient and viable tool for researchers.

A Case Study of the Performance of Speculative Asynchronous Simulation on Parallel Computers. PATRICK WILKERSON (Austin Peay State University, Clarksville, TN, 37044) KALYAN PERUMALLA (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Modern supercomputers use thousands of processors running in parallel to achieve their high computational speeds. However, on such large processor counts, communication and synchronization operations can waste valuable processor time. Communication involves processors exchanging intermediate computed data that needs to be shared by processors at runtime. Synchronization involves processors ensuring the mutual orderings of operations across processors are correct. In this work, we investigated the runtime efficiency of two methods that are aimed at reducing communication and synchronization costs, respectively, namely, asynchronous updates and speculative execution. The experimental investigation is performed on a parallel finite difference time domain (FDTD) simulation developed at ORNL, which has wide applicability in simulating various physical system phenomena. It uses an iterative algorithm to reduce communication by allowing messages to be asynchronously sent when the change in values on a given processor is greater than some threshold value. We conducted research to develop an empirical performance study of the algorithm. The first part of asynchronous updates is accomplished by exploring the effect of threshold-based communication on overall runtime of the parallel simulation, with the number of processors increasing. A significant improvement in performance on up to 64 processors was observed when using the asynchronous update scheme, due to reduced communication. We are currently working on the synchronization aspect of the problem, to relieve the tight coupling among processors, using "speculative execution" with sophisticated "rollback" techniques being developed in an ORNL Laboratory Directed Research and Development (LDRD) project. In speculative execution, processors are allowed to progress without having to wait for other processors, but any violations in ordering of computations are detected and corrected using rollback techniques. Further research is being done to implement a rollback mechanism necessary for optimized parallel execution. Preliminary results on a special case of the synchronization mechanism show improved speedup over and above the gains of asynchronous updates on up to 64 processors. We are working towards implementing the more general rollback method and expect to complete the performance study on the generalized asynchronous speculative execution.

A Computational Process to Locate IS elements and Study Horizontal Gene Transfer in Bacterial Genomes. WALTER LEWIS (Cheyney University of Pennsylvania, Cheyney, PA, 19319) SEAN MCCORKLE (Brookhaven National Laboratory, Upton, NY, 11973)

Currently there is a tremendous focus on studying how whole units of DNA are shared between species. These movable tracks of DNA, called transposons, are demarcated by shorter insertion sequence (IS) elements, which themselves are generally around 700 to 2500 bp in length, and code for proteins implicated in the transposition activity, such as transposase and intergrase. IS elements are bracketed by pairs of inverted repeats of variant length (approx. 10 -30 bp), and are further embedded in a pair of short direct repeats of approximately 3 – 7 bp. While there is a great deal of software to automatically annotate genes and regulatory regions, there are currently no programs to identify IS elements de novo. Hundreds of completed bacterial genomes are currently available and more are constantly being added to that list, which makes it increasingly important for the computational detection of IS elements. To this end, we have developed a C++ program to locate IS elements in bacterial genomes. The process was broken down into steps in order to make it more dynamic so that one could easily examine the data in each stage of the process to detect errors, or perhaps discover something unexpected. First, all repeats (within length constraints) are reported, then these are searched for pairs of inverted repeats which have the proper orientation and distance. Positions, paired inverted repeat sequences, and interior coding sequences are reported for these IS element candidates. Lastly the interior coding sequences are examined for homology with known transposition genes. Preliminary tests on the genome of Enterobacter strain 638, recently sequenced as part of the DOE biofuels initiative, revealed 9 putative IS candidates, 6 of which were confirmed by Blastx searches of Genbank, which yielded strong homology matches to transposase and intergrase in other organisms. Further testing on an IS-rich genome, now Cupriavidus metallidurans (CH34), will allow us to optimize program parameters to improve performance. Because this process reports direct and inverted sequence pairs, which travel along with the IS element as it moves, it can be used to trace horizontal gene transfer history within or between genomes for evolutionary studies.

A New Build System for The Common Component Architecture. DANIEL TAYLOR (Edinboro University of Pennsylvania, Edinboro, PA, 16444) BOYANA NORRIS (Argonne National Laboratory, Argonne, IL, 60439)

Consolidating Three Data Sources Into a Centralized Data Repository That Can be Accessed by Multiple Systems. TASHA WHEATLEY (Clarion University, Clarion, PA, 16214) MARK BAYLESS (Pacific Northwest National Laboratory, Richland, WA, 99352)

The purpose of the system is to pull data from three different Lotus Domino application databases, Office of Science Application System (Edulink), SEE Application System (OFP), and SEE Request System (SEE), into a centralized location. The Edulink system provides and controls all online applications filed for DOE internships at any of the national laboratories while OFP merely contains applications for internal Pacific Northwest National Laboratory (PNNL) internships. In addition the SEE system allows PNNL mentors to request an intern that applied for an internship through either the Edulink or OFP systems. Before being pulled into the private Intern Tracking System database, the original system used the SEE System to pull information from the Edulink and OFP systems. Because of complexities in how data was being pulled together by the SEE system from the Edulink and OFP systems, modifications were needed to address data integrity issues. The new design provides a Domino agent that executes daily to compile the necessary data from the three databases. Once the data is compiled, a Microsoft SQL Server scheduled job runs to load the data from Domino into SQL Server tables, providing any final data scrubbing such as converting dates in varchar format to date/time format. A SQL Server view was also created so the Data Repository (Op Warehouse) can pull and store the data; thus, allowing several internal systems to access the data from a central source. This is ideal because it ensures consistent data among the systems. In the next couple of years the applications are going to be moved from Domino to a .NET application, which is another reason why these modifications needed to occur. Whenever the new applications are created, the hope is to ensure that all necessary information is available for all systems.

Converting RELAP5-3D Environmental Library to Fortran 90. RILEY CUMBERLAND (North Carolina State University, Raleigh, NC, 27695) DR. GEORGE MESINA (Idaho National Laboratory, Idaho Falls, ID, 83415)

RELAP5-3D is being converted to Fortran 90 to increase code legibility and longevity. As part of that effort, the RELAP5 subdirectory called the Environmental Library (EL) has been converted to Fortran 90 with the additional requirement to increase code readability and conciseness. The EL is comprised of service subroutines that perform table look-ups, interpolations, and solutions to linear systems, among other things. To improve readability, each EL subroutine was initially restructured the commercial tool, For_Struct, which applies uniform coding style rules and reorganizes to code to eliminate jumps in logic flow paths (GO TO statements). Afterwards, manual modifications were made to further improve readability. These modifications included eliminating multiple returns, entry points and virtually every GO TO statement. Pre-compiler directives also reduce legibility; these were eliminated manually by replacing machine-specific code with machine-independent Fortran 90 intrinsics and by moving other directives into Fortran 90 modules or new subroutines. The files were further processed as groups by abstracting coding repeated in several subroutines into separate, reusable subroutines. An example of this was the many search loops reduced to two subroutines. In addition, a mnemonic subscripting system was used in many subroutines and transmitted through a new Fortran 90 module. Another improvement was rewriting the subroutines from older fixed format Fortran to the more modern free-source format. To achieve this, another pre-compiler that alters the code for 32-bit or 64-bit integer machines was eliminated by creating and using a bit-transfer subroutine of Fortran 90 constructs. Additionally, handling the declaration with a kind parameter set in a module by formatting all floating point constants as double precision was also necessary to convert to free format. The three multi-file processes improved conciseness. Measurements of improved readability and conciseness were made. The conversion process resulted in a sizable reduction in the number of line labels and GO TO statements, an indicator of readability. The number of pre-compiler directives was reduced dramatically, improving conciseness and readability. The number of lines of commented lines increased slightly while the number of non comment lines, an indicator of conciseness, decreased slightly. This process has clearly improved the readability of the code.

Cyber Security at Oak Ridge National Laboratory. BOBBY SWEET (Roane State Community College, Oak Ridge, Tn, 37840) VICTO HAZLEWOOD (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

The Office of the Chief Information Officer (OCIO) has the responsibility to ensure that all information technology (IT) is managed to all statutory, regulatory, and organizational requirements which can be found in the Certification and Accreditation package which documents the Oak Ridge National Laboratory (ORNL) Cyber Security Program Plan. In the case of ORNL, who has over ten thousand computer systems, there are systems that will not comply with these standards for a variety of operational and other reasons. The OCIO ensures that computing resources are managed to requirements with some mitigation in place for systems where the requirements could not be met. In many cases, a security plan will provide the basis for the mitigation for systems or collections of systems that cannot meet all the requirements. Security planning was our main emphasis this summer. There were approximately two-hundred forty systems at the start of this summer that needed a security plan in place, if they were going to stay on the ORNL network. A process needed to be developed and implemented to address the two-hundred forty systems and their required security plans in the most efficient manner within a limited amount of time. Four security planning meetings were given to communicate the security planning process. It is expected that these meetings will be held each month to keep up with new systems that might not meet requirements and require a security plan. Three questionnaires were created for the different control categories to help the owners of the devices. The three different control categories were moderate with enhanced, moderate, and low. The categories are determined by the sensitivity of information each of these systems would hold (moderate with enhanced requiring the most controls of the three). The classes have been completed and the resources have been distributed to the system owners and administrators to start their security plan process. The ORNL security planning project has been developed and is being implemented. The security planning process will continue to be improved in the effort to protect the devices that are not able to satisfy all cyber security requirements. This project is designed to protect the devices making them as safe as possible from intruders or unauthorized access by providing a mitigation strategy when a device cannot implement one or more required cyber security requirements.

Data Analysis and Social Collaboration. EUGENE SANDBERG (Mississippi State University, Starville, MS, 39762) DANE SKOW (Argonne National Laboratory, Argonne, IL, 60439)

Developing a Proof-of-Concept CyberInfrastructure for the USA National Phenology Network. BENJAMIN CROM (University of Tennessee, Knoxville, TN, 37916) BRUCE E. WILSON (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Phenology is the study of the timing of naturally recurring events, such as flower blooming, particularly those influenced by environmental changes. Phenological measurements are extremely valuable indicators of local, regional, and global climate changes. Further, phenological variations have a variety of crucial environmental and socio-economic implications. Despite this value, broadly distributed phenological measurements are almost non-existent in the USA, in contrast to some European countries. The objective of this work is to explore and evaluate different data models and data management methods needed for the development of the cyberinfrastructure for a new data resource: the USA National Phenology Network (USA-NPN). The USA-NPN will create a national repository for submitting and retrieving a wide range of phenological data, including data from existing research networks and from citizen scientist observers. For this project, we have been focusing on the data model and information management tools necessary to support the citizen scientist aspect of USA-NPN, based on existing work from the Plant Phenology Network (PPN). A key feature at this stage of the program is for citizen scientists to enter phenological observations on any of the 38 species currently in the PPN. The USA-NPN will also require basic tools for observers to retrieve their own data and the publicly available data. We have created a simple but flexible database schema which we are testing against a range of ways the system may be used. This schema is an improvement over the current PPN tools, which use separate text files for each species. We are also working on improving the registration and authentication processes. The user is now able to create login credentials and submit personal contact information without PPN staff intervention. This reduces staff load and the revised data entry forms greatly improve the tracking of data provenance. Our current work will improve session management operations and may enable OpenID authentication. We intend to streamline the maintenance of data submission pages by eliminating the current 38 separate submission pages in favor of one master submission page driven by the underlying database. We will also add a data retrieval page which will give users access to public phenology data. The data offered by this network will contribute to a better understanding of climate change and help engage citizen scientists in climate change science.

Developing a Tree Search Algorithm for Track Reconstruction in the Hall A BigBite Spectrometer. BRANDON BELEW (Rensselaer Polytechnic Institute, Troy, NY, 12180) OLE HANSEN (Thomas Jefferson National Accelerator Facility, Newport News, VA, 23606)

In many experiments that involves particle collisions, the ability to track the paths of the particles is essential. In the Hall A BigBite Spectrometer at the Thomas Jefferson National Accelerator Facility (JLab), wire drift chambers are used to detect these paths. For each individual measurement, there is an associated hit pattern that shows which detector bins were activated by the particles traveling through the drift chamber. The track information can be reconstructed from the hit pattern with the help of software. The current algorithm uses a brute-force method to analyze all possible ways of connecting lines through the hits. This has been sufficient thus far, but future experiments will need faster performance because the expected rates and particle multiplicities will be higher and a larger number of wire planes will be used. The HERMES Spectrometer at the Deutsches Elektronen-Synchrotron uses a much more efficient tree search algorithm. The goal of this project was to implement such an algorithm for use at JLab. The code developed for this project was written based on a general description of the HERMES algorithm. It compares the associated hit pattern to a pre-generated database of all possible patterns that could be produced by particles moving in a straight line, arranged in a branching structure of increasing resolution. This tree structure allows for logarithmic rather than linear comparison time. It has been demonstrated to work with a high level of accuracy using Monte Carlo simulations. For a sufficiently high detector resolution and reasonably low levels of simulated background noise, the algorithm was shown to very quickly match all the simulated tracks and to produce very few “ghost tracks,” or false positives. In the future, this algorithm will be moved out of the testing phase and integrated with JLab's existing software to be used on a regular basis. The increased speed of this new matching algorithm will make many future experiments possible.

Developing standardized interactions between a variety of data sources and generally available map-display programs. WILLIAM KARNESKY (Columbia Basin College, Pasco, WA, 99301) DAVID MILLARD (Pacific Northwest National Laboratory, Richland, WA, 99352)

There is a wide variety of free service-based map programs aimed at the general user such as Google Earth, Google Maps, Microsoft’s Virtual Earth, and NASA’s World Wind. These map displays allow a user to add their own points of interest in a variety of ways including place-marks, polygons, and image overlays. There are many tools and applications that could benefit from the ability to connect their capabilities with a service-based map display. One example would be to dynamically link an Excel spreadsheet containing the location of a county’s traffic control points to Google Earth. Currently there are very few generic tools that provide efficient and standardized ways to dynamically link an application’s information with a service-based mapping program. This project’s purpose is to identify and develop mechanisms that make it simple to link map display capabilities to an external application. PNNL’s status board system called WebOSB which provides the dynamic status of user-defined information such as bridges, emergency shelters, traffic control points, etc. was used to develop an interoperability layer that will link data sources containing status information to a service-based map display capability. After a map-display interoperability layer is implemented in WebOSB, PNNL plans to continue to generalize that capability to allow a quick, seamless integration of data-oriented systems with a map display capability.

Development of a Model for Electrical Distribution Utility Switching Analysis. SHAWN WALTON (Big Bend Community College, Moses Lake, WA, 98837) CLINT GERKENSMEYER (Pacific Northwest National Laboratory, Richland, WA, 99352)

The ability to assess and respond to a situation is imperative in the operation of an electric utility. Electric utility dispatchers are required to reconfigure the system on a daily basis, but may not always have the necessary instructions or data to make correct decisions. The goal of this project was to create a Model for Electrical Distribution Utility Switching Analysis (MEDUSA). The model was created using Invensys Wonderware, Practical Extraction and Report Language (PERL) scripts, and a Structured Query Language (SQL) database. PERL and SQL were used to create simulated electrical line values, including amps and volts, for two distribution lines. Using Wonderware, a graphical user interface was created to represent the distribution line system including breakers, loads, connecting lines, and switches. This interface allows the dispatcher to manipulate any combination of breakers and switches and view the resulting configuration and estimated system measurements. Once the configuration is simulated and optimized, the switches can either be operated directly from the MEDUSA dashboard or instructions can be given to a lineman to operate them manually. All possible breaker/switch configurations were tested and proved. The graphical interface was designed to be visually clutter-free and utilized color schemes native to the industry. MEDUSA demonstrates the effectiveness of providing situational awareness to distribution system dispatchers by allowing them to simulate the new switch configuration and analyze the resulting data in real time without risk to the system. Further development of this program could benefit from eliminating the current Wonderware shell and utilizing the C++ language. This would improve the ability to easily modify or add to the system. Integration of actual system measurements and the modeling of an entire distribution system would also need to be addressed for this application to be utilized by the industry.

DHS Analyist Software Implimentation. CHRISTOPHER WHITEHOUSE (Big Bend Community College, Moses Lake, WA, 98837) MARY SUE HOXIE (Pacific Northwest National Laboratory, Richland, WA, 99352)

The objective is to see how analytical programs such as Collaborative Analytical Tool (CAT), ProSPECT, IN-SPIRE, and Analyst Notebook (ANB) will help in analyzing investigative data. Before any of the tools could be used we needed to convert the data into a usable format, plain text. My lab partner, Sherwin Hunte, created a program to automate extraction of a PDF (portable document format) file into separate smaller PDF files. After converting all the documents we imported the data into the separate programs. I used CAT to index all of the files related to the data. These files included Excel, Access, Adobe Acrobat, and text documents. CAT is very useful because it can index several different file types and group them into different data sets. I also imported the data into the ANB program and created timelines for the data. This was a very useful tool except after importing the data it requires a lot of the organizing of the timeline by hand. These analytical programs simplify looking at data that is convoluted. They allow us to visually search through and cross reference the data. These programs also make it a lot easier to find specific information and reference it as needed to support or refute Hypotheses, which allows for stronger hypotheses and better understanding of the data being analyzed. Though the way the data is gathered may need revising, I do think these tools would be helpful to any analyst working with convoluted or disjoined data.

Electrodymanics Simulation of the Pulse Line Ion Accelerator. ROXANNE MARTINEZ (Stanford University, Stanford, CA, 94305) ENRIQUE HENESTROZA (Lawrence Berkeley National Laboratory, Berkley, CA, 94720)

The Heavy Ion Fusion Science Virtual National Laboratory is currently studying the Pulse Line Ion Accelerator (PLIA), a slow-wave accelerator that would substantially reduce the cost of experiments for studies of high energy density matter and fusion ignition conditions. The PLIA uses a high voltage pulse at the input end of a helical pulse line structure as a means to accelerate charged particles. The high voltage pulse generates an electric field that travels across the span of the accelerator. This wave pulse can be used to accelerate the ions to energies far greater than the peak voltage which was applied to the input end of the structure. Moreover, the PLIA can axially confine the heavy ion beam bunch, making it an excellent candidate for a high intensity, short bunch injector. The electrodynamics of the system will be presented. This analysis was performed using MAFIA (MAxwell’s equations by the Finite Integration Algorithm), an electromagnetic and particle dynamics code.

Energy Efficiency of Desktop and Laptop Computers. MAT DOIRON (Big Bend Community College, Moses Lake, Wa, 98823) CHARLIE VERBOOM (Lawrence Berkeley National Laboratory, Berkley, CA, 94720)

As the number of computers in use worldwide increases, so does the demand for electricity to power them. This research focused on measuring computers typically used at Lawrence Berkeley National Laboratory in order to determine their power consumption. This included the Optiplex 745 series from Dell, ClientPro series from MPC and a custom made computer designed to run Computer-Aided Design (CAD) software. Three Dell Latitude Laptops( D600, D610, D630) were included. Measurements were taken during five different “states”; off, booting, idle, active and sleep, with a Kill-A-Watt meter. Two Operating systems, Microsoft Windows XP Professional and Windows Vista Business were also compared to determine their impact on power use. PassMark Performance Test V 6.1, a benchmarking program, was used to evaluate the performance of the computers. It was also used to simulate the computer in an active state. It was discovered that the computer uses less energy while sitting idle than while active and significantly less while in sleep mode. Putting the computer to sleep during non-working hours would be ideal but several factors currently prevent this from taking place, including scheduled backups and remote access. New desktop computers being implemented at the lab use less energy than those currently in use despite their superior performance. The new computers are the Dell Optiplex 745 series, which were tested in three different configurations: Minitower, Small Form Factor and Ultra Small Form Factor. The three were compared to see if size or component differences affected power consumption and performance. There was very little difference between the Minitower and Small Form Factor. In contrast, the Ultra Small Form Factor used less power with reduced performance. This could be because of the external power brick and lack of dedicated graphics card. The laptops in this study consumed significantly less power than did the desktops: 22W while idle compared to 68W. This is significant because it includes the monitor. An LCD monitor uses 30W and a CRT up to 130W, which is more than the entire laptop. Windows Vista and Windows XP were also compared using identical computers. Vista performed worse and consumed more power despite the fact that it is the successor to XP. It is recommended that customers wait for the first service pack to switch to Vista if performance or power management is a major concern.

Enhancing Network Security by a Centralized and Standardized Dynamic Event Log Data Collection Method. MARK SEAL (Walters State Community College, Morristown, TN, 37813) DON WILLIAMS (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Event logs are at the core of network security. Without the tracking and logging ability that event logs allow, there would be no accountability of the security of the network. The logs allow the detection of worms and viruses via monitoring system resources and processes, detection of port scans by monitoring the dropped port requests by the firewall, detection of brute force password cracking by the number of failed login events, and the ability to start a trace on the agent when a security event is recorded. It is imperative that the logs from each machine are parsed and the relevant information passed on to a central log server (syslog) where the information can be analyzed and trend analysis preformed. The problem faced is that Windows does not have a single log that collects all of the required data; instead the data must be collected in segments. In addition, both XP and Vista are both being used as operating systems. Each of these OS’s has different ways of storing the logs, requiring different data collection methods. XP uses evt format as the logging method while Vista uses an evtx format. XP logs may be ported to the syslog server while Vista requires that either an event viewer or a data collection program (e.g. LogParser) be used. The development of the new Windows event log data collection protocol centers around three core concepts: scalability, control, uniformity. The chosen method of collection of data by using Powershell scripts meets all three core concepts. Powershell may be installed on both XP and Vista machines, while coming mounted on the upcoming Window Server 2008. Scripting in Powershell allows for the full control of the event log data. The data can be parsed by content as well as being stripped of superfluous data for storage and trending in the syslog server. In addition, the data will be supplied to the syslog in real-time, allowing for real-time trend analysis and alerting protocols. Powershell scripting also creates a uniform event log data collection protocol as that Linux system logs are parsed by scripts. Powershell, which is based on the Korn Shell model, will be easier for cyber security to modify and maintain the core scripting code. This project will increase the security of the network by allowing for data collection of the system logs of Windows machines while decreasing the response time of the cyber-security team by allowing for real-time data comparison between Linux and Window machines.

Event Based NeXus Files. WESLEY KENDALL (University of Tennessee, Knoxville, TN, 37849) PETER PETERSON (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

When gathering data from instruments across different locations, a universal file format for representing the data becomes essential. The NeXus format is a standard for representing raw data. My project involves writing a program to convert neutron event files to the standard NeXus format. The main resource to write this software is the NeXus Application Programming Interface (API). The API includes many functions for reading and writing data in the NeXus format. Currently, raw binary neutron event data is automatically converted into histogrammed NeXus files. This locks the scientist into a particular choice of histogramming parameters for the data. If the scientist wishes to perform a histogram parameter data analysis scan to find the optimal settings, a new histogram file for each parameter set needs to be generated before data analysis can proceed. An event NeXus file will allow a single file to become the source for each set in the parameter data analysis scan. Space will be significantly reduced since the original raw binary neutron data and its associated metadata will be duplicated in the event NeXus files. My program converts the raw binary neutron event data into event based NeXus files enabling the previously described flexibility. The following details further illustrate how my program operates. First, it reads in a neutron event file, which consists of pixel identification (id) numbers and pixel time of flight values. Since some detectors have different electrical set ups, the pixel id numbers will often have to be mapped to other numbers to closer match the physical detector layout. After the pixel id numbers are mapped to the appropriate number, the program then reads in a pulse id file. The file gives the exact starting times of a pulse. Adding the time of flight value of a pixel id with the corresponding pulse time gives the exact clock time of the neutron event in nanoseconds. This allows for analysis of different time slices of data. After all the steps are completed, the event based NeXus file is created. The data can then be analyzed by scientific software that is specifically made to handle the NeXus format, making analysis much more efficient.

Extending the MPI-Serial Library to Support Derived Datatypes. JOHN YACKOVICH (Shippensburg University of Pennsylvania, Shippensburg, PA, 17257) ROBERT JACOB (Argonne National Laboratory, Argonne, IL, 60439)

Face Recognition System. AMARACHUKWU ENEMUO (The City College of New York, New York, NY, 10031) DR. UPENDRA ROHATGI (Brookhaven National Laboratory, Upton, NY, 11973)

Biometrics, a significant component of identification technology, involves verification and identification based on an individual’s biological characteristics. Universality, time invariance, uniqueness, accuracy, and acceptability are desirable characteristics biometrics must possess. Commercial biometric systems are designed to use a single biometric; however, the Integrated Biometric Recognition System (IBRS) consisting of fingerprint, voice, face, and hand Recognition Systems (RS) combines all four technologies into one system, and is platform independent. The IBRS utilizes client-server architecture, and was scheduled to be tested to determine the efficiency of its performance. To fully assess this, understanding of the underlying algorithms are essential. IBRS has three modes namely Enrollment, Identification and Verification. For face recognition, prior to the Enrollment phase, a sample of multiple face images are collected from various individuals. These set of face images become our database of known faces. (S) IBRS uses the eigenface (eigenvector) technique to recognize the face image of a client. Eigenfaces are based on the Euclidean distances of salient features on the face. These distances are measured, and stored as general templates which are used for identification of the face area on an image, and for verification. During Enrollment, a camera is used to capture fifteen sequential images of a client which are converted to grayscale images of same size in pixels. Each pixel is converted to a vector, and 15 eigenfaces are constructed from the face images of the client and stored as the client’s personal template. The Identification phase, involves a 1 : S comparison where "1" represents the client’s face image and S is the database of known faces. Verification is a one-to-one authentication of who the client portrays himself to be. During this mode, the client’s input image is compared with the stored personal template created at the enrollment stage. Face testing on IBRS was done using frontal view of face images and varied lightning. It was found out that the system is easy to use and the Enrollment, Identification and Verification modes work as expected.

FitNesse: Application of a Software Testing Framework to the Bioinformatics Resource Manager. JONATHAN RICE (University of Nebraska-Omaha, Omaha, NE, 68182) ANUJ SHAH (Pacific Northwest National Laboratory, Richland, WA, 99352)

One of the goals of a software development team is the delivery of quality software in a timely fashion. The importance of thoroughly testing a software product can never be overstated. However, testing the functionality of software, especially ones with graphical user interfaces (GUIs), is often a laborious process due to the amount of time required, difficulty of testing consistently, uncertainty of what exactly should be tested, and the lack of easy-to-use testing frameworks. Other considerations include minimal impact on existing code, a straightforward script writing process, and automated execution of test suites. The project goal was to integrate an existing automated testing framework into the Bioinformatics Resource (BRM) application to facilitate the automated testing of key functionality. Previously, testing was manual and most functionality testing occurred just prior to version releases, rather than progressively. Project development utilized the FitNesse testing tool, which allows for the creation and running of acceptance tests against GUI-based software; BRM; and the Java programming language. The ease of use, the script-based test case development and the WIKI-like execution engine make FitNesse amenable to use by non-developers. The new test environment allows convenient, consistent, and defined testing.

Flocking-based Document Clustering on the Graphics Processing Unit. JESSE ST.CHARLES (University of Tennessee at Chattanooga, Chattanooga, TN, 37415) XIAOHUI CUI (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Analyzing and grouping documents by content is a complex problem. One explored method of solving this problem borrows from nature, imitating the flocking behavior of birds. Each bird represents a single document and flies toward other documents that are similar to it. One limitation of this method of document clustering is its complexity O(n2). As the number of documents grows, it becomes increasingly difficult to receive results in a reasonable amount of time. However, flocking behavior, along with most naturally inspired algorithms such as ant colony optimization and particle swarm optimization, are highly parallel and have found increased performance on expensive cluster computers. In the last few years, the graphics processing unit (GPU) has received attention for its ability to solve highly-parallel and semi-parallel problems much faster than the traditional sequential processor. Some applications see a huge increase in performance on this new platform. The cost of these high-performance devices is also marginal when compared with the price of cluster machines. In this paper, we have conducted research to exploit this architecture and apply its strengths to the document flocking problem. Our results highlight the potential benefit the GPU brings to all naturally inspired algorithms. Using the CUDA platform from NIVIDA®, we developed a document flocking implementation to be run on the NIVIDA® GEFORCE 8800. Additionally, we developed a similar but sequential implementation of the same algorithm to be run on a desktop CPU. We tested the performance of each on groups of news articles ranging in size from 200 to 3000 documents. The results of these tests were very significant. Performance gains ranged from three to nearly five times improvement of the GPU over the CPU implementation. This dramatic improvement in runtime makes the GPU a potentially revolutionary platform for document clustering algorithms.

FORMS-GUI: A GRAPHICAL USER INTERFACE FOR INPUT INTO THE FORMS SIMULATION TOOL. SEAN DOEBLER (Drake University, Des Moines, IA, 50311) DR. BOYANA NORRIS (Argonne National Laboratory, Argonne, IL, 60439)

Georeferenced Image Registration of Aerial Imagery. JASON KOTENKO (Edinboro University of Pennsylvania, Edinboro, PA, 16444) RYAN HRUSKA (Idaho National Laboratory, Idaho Falls, ID, 83415)

Photography captured from unmanned aerial vehicles (UAVs) can be used in a variety of applications ranging from security to thematic analysis of geographic areas. Images taken from an aerial platform of the earth’s surface must be overlaid to form a single mosaic, this is called image registration. Georeferencing is the process of accurately projecting the images onto a common reference frame. To accomplish this, invariant features were extracted from each image. The search space for matching these features between images was narrowed through the analysis of telemetry data about the location and attitude of the UAV when the picture was taken.

Implementing Optimized and Scalable Non-Standard Matrix-Multiply Algorithms. RANDY HASSELL JR. (Old Dominion University, Norfolk, VA, 23455) RICHARD BARRETT (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Matrix-Multiplication is the fundamental operation for solving several algorithms including other linear algebra operations like matrix inversion, systems of linear equations, and matrix determinants. It’s also used in computer graphics and digital signal processing. The Computational Materials Science Group at the Oak Ridge National Laboratory (ORNL) is developing a large-scale computer application that studies the fundamental properties of materials. The performance of this materials science code relies heavily upon the execution of matrix-multiplication. The dimension of the matrices are 32 x N, where N is currently 4K, and continues to expand to 40K. The purpose of this project is to create an optimal algorithm that performs matrix-multiplication, and to research the implementation on different computer architectures. Using the Fortran language, all the matrices are dynamically allocated, so that memory is created and destroyed without wasting any. Then, double precision random numbers fill both matrices. There are three multiplication techniques to test, a Fortran function matmul, a source file dgemm.f, and the subroutine dgemm directly from lblas during compilation. The timer used is a subroutine called gettimeofday() in the C language. The algorithm calculates the Floating Point Operations per second (FLOPS) that measure an algorithm’s performance on a processor. For multiplying rectangular matrices, I am testing from 5x5 to 40Kx40K using iterations and 10 trials each. For multiplying square matrices, I am testing from 5x5 to 4Kx4K. I am compiling my algorithm with these computer processors: Intel Celeron, Pentium 4, and the Cray XD1(dmz00 & wc00). For matmul, the performance dropped as the size grew from an average of 5GFLOPS to 1 MFLOP on all processors. For dgemm.f, the dmz00 performed the best leveling around 225 MFLOPS as the size increased. For square matrices, the performance dropped as the size grew on all processors, and that is due to the increased amount of time it takes to multiply square matrices because after all, a 1Kx32 takes less time to read than a 1Kx1K. Overall, the best performance came from the wc00 using dgemm, where the performance leveled at 600 MFLOPS for the larger matrices. The implementation of the algorithm will continue on the hardware side by taking into account the layout of the memory and the transport speed from the main memory to the local cache, and then the materials code should run optimally.

Java Table Tools for the Collider-Accelerator Department. DAVID KATZ (Alfred State College, Alfred, NY, 14802) SETH NEMESURE (Brookhaven National Laboratory, Upton, NY, 11973)

The Collider-Accelerator Department at Brookhaven National Laboratory has a diverse set of data tables which are used to display accelerator controls system data. One program used to display the data is called “pet” (Parameter Editing Tool). There are also a mature set of database tables that store various types of information including device description data, logged data, configuration data and more. Java tools to display the data for this complex system would be very beneficial since Java is platform independent and will be able to run on any OS in the lab. The tools include a unique table structure that doesn't have dependencies on the data source (i.e. database or controls system). In order to provide this in the implementation, I looked at the different types of control system parameters and made the decision to use a hash mapped based table structure. These maps store everything from the look and feel to the layout of the data within the cells. A wrapper class contains each of these hash map arrays, thus defining the generic table structure. Two separate parsers were also written, one to parse pet files and the other to parse database tables. Both of these classes return a table structure. In order to display the table, a Java table model, cell editor, and cell renderer were written to use the table structure with its unique key mappings. This design offers flexibility for future implementations. The intention is that this tool will allow a sophisticated interface to be quickly and easily integrated by programmers into end-user applications while also providing a clear interface for the user.

Long Island’s Interactive Weather Map and Microsoft Excel Data Entry. JOE LANIER (State University of New York at Stony Brook, Stony Brook, NY, 11790) VICTOR CASSELLA (Brookhaven National Laboratory, Upton, NY, 11973)

The Long Island Interactive Weather Map was created for all personnel on site at Brookhaven National Laboratory (BNL). The primary purpose of this map is to give employees a place where they can access current weather data from all over Long Island, as well as archival data dating back from the 1940s up to the present day. Throughout weather stations across Long Island, we collected data on temperature, wind speed, direction, gusts, barometric pressure, humidity, and precipitation from old records and entered them into a large Microsoft Excel database. This data is sent via radio waves to a receiver at BNL, collected and stored as an ASCII string, fed into a MySQL database, and updated on the meteorological website once a minute. We have created a web page using the following languages: PHP, SQL, JavaScript, and HTML. This page has an image of Long Island and displays all of the towns with BNL’s current weather stations. When a user hovers over any of the towns or stations, a small pop-up box appears and the current data is displayed for that town. At the bottom of the box, a link for further archive data is available for that town or station. The user can navigate through the data to get what is needed. Never before has there been such an easy, all-in-one design that has allowed users to see current and archival weather data on BNL’s meteorological website. Throughout this project, the majority of the BNL weather acquisition system has been reconfigured and available to the BNL staff.

Mass Metabolic Model Generation (3MG) in the SEED. KEVIN FORMSMA (Hope College, Holland, MI, 49423) MIKE KUBAL (Argonne National Laboratory, Argonne, IL, 60439)

OSCAR Diskless. WILLIAM ADERHOLDT (Tennessee Technological University, Cookeville, TN, 38505) STEPHEN L. SCOTT (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

OSCAR, Open Source Cluster Application Resources, is software used to create and manage high performance computing clusters. The current method of installation assumes that each compute node is equipped with a local hard disk. However, the trend in high performance computing today is leaning towards diskless clusters where compute nodes do not require local hard disks. The motivation for diskless clusters include reduced power, lower hardware costs, and improved reliability (hard disks are notoriously prone to failure). This work extends OSCAR to support the creation and management of both diskfull and diskless clusters. In order to enable diskless OSCAR nodes, a boot method known as NFS-Root is used. This means that during the bootstrap phase the compute nodes will mount their root partition via a Network File System (NFS) share, eliminating the local hard disk requirement. The way this is done, specifically, with OSCAR is to mount a small root partition containing only what is absolutely needed for the init process as read/write. All other directories are mounted as read-only from a NFS shared network folder, which is accessed by the diskless nodes in the cluster. Init is modified in order to obtain a diskless node by mounting the network shares before any other work is done. So, through creation of a NFS shared network folder, a small root partition, and modification of the init process, OSCAR is able to create and maintain both diskfull and diskless clusters. This ability aligns OSCAR with the current trends in high performance computing and creates more efficient clusters as it reduces power, lowers hardware costs, and increases reliability.

PDF Forensics and Steganography Detection. BRANDON LANGLEY (University of Tennessee, Knoxville, TN, 37996) MIKE NEERGAARD (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

The Portable Document Format (PDF) is used to display two-dimensional documents in a device-independent manner. Standard PDF viewer applications, such as Adobe Acrobat, are designed to hide more information about PDF files than they display. A PDF file consists of a series of objects that are referenced whenever they are used. There are various methods of hiding content inside a PDF file so that a viewer will not display it. This makes PDF an ideal format for hiding information, because most people will view the file in a standard viewer and will not dig deeper into the structure of the file. The science of hiding this information so that only certain people know there is anything hidden is known as steganography. There is no reason for a person to go out of their way to dig into a stegonagraphic document, because nothing appears hidden to the normal eye. I am researching the different ways to hide content inside a PDF file and, in turn, the different methods of extracting hidden data from a PDF file. I am also writing a standard C++ Linux application package to aid in the extraction of any hidden data present in the file. Currently, there is no application designed to help with PDF forensics in this manner. I am using the Adobe PDF Reference (second edition) to study the complete PDF specifications, and the Poppler general public license library to help with the PDF parsing. The application first prints all metadata from the file that may be useful for a forensics investigation. This includes the author, application used to create the file, creation date, modification date, etc. In addition, this application searches for and displays any content that is hidden in the file and not viewable in a standard PDF viewer. This is a three phase process. First, the application finds every legal object that is in the file but never referenced by a page. Second, it searches for any content that is hidden between legal objects. Finally, the application looks at the instructions for displaying each page. If an object is referenced by a page, but never used in the page’s actual display instructions, those objects are also considered hidden. Any hidden content is then displayed for the user. This application should aid in investigations involving computer crimes such as child pornography. If someone has criminal content hidden in a PDF file so that only they know how to view it, this application should help to expose it.

Penetration Testing of ORNLGOV. TRISTAN BITHELL (University of North Carolina at Charlotte, Charlotte, NC, 28078) DON WILLIAMS (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Today, the risk of an internet based attack is greater than ever. Google.com has become the target of hackers to be used as a tool for reconnaissance. Google is capable of revealing vulnerabilities in web sites if you know the proper search operators. Through the use of a Google API (Application Programming Interface) key one can even automate the reconnaissance process. Other tools such as spider crawlers and backend tools can also reveal vulnerable files or URL’s that should have remained hidden. Also, the listing of excessive contact data can be used in a social engineering based attack. My project was to use manual searching and automated tools to attempt to locate any vulnerabilities or unprotected, sensitive data on onrl.gov or any leaked information from the lab to a newsgroup. I used Athena 2.0, Wikto, Paros, SiteDigger 2.0, and Aura in my project. During my initial setup, I found that Google no longer issues API keys so I was able to use SnakeOil Lab’s Aura program to simulate an API key. I used Athena 2.0, Wikto, and Site Digger 2.0 to look for common vulnerabilities that were listed in the Google Hack Data Base at johnny.ihackstuff.com. Wikto is a multipurpose tool that I used for several tests. First I scanned the website for known vulnerable directories. Next I scanned those directories for known vulnerable files. Wikto would use an AI to assess if a URL or file existed but this tended to often return false positives and lead to URL’s that did not exist. With Paros I set up my browser to redirect traffic through a local proxy server run by Paros to intercept and modify data packets. As I browsed ornl.gov, Paros would save the URL’s that I visited. I then activated a spider scan on a URL to generate a list of every directory and file in the website. Though ornl.gov itself proved too massive to scan in its entirety, I was able to complete a scan of neutrons.ornl.gov which is the website for the SNS facility. Through my penetration testing, I was able to extract various pieces of possibly sensitive information. From the results of the tests and measures taken to secure ornl.gov’s data, the website will be more secure against the threat of an attack.

Performance Comparison between 10GigE and InfiniBand in high performance computing. COLBY BOYER (University of California, Berkeley, Berkeley, CA, 94720) CAROLYN M. CONNOR (Los Alamos National Laboratory, Los Alamos, NM, 87545)

The processing power of a large computing cluster relies heavily on its ability to transmit data quickly among its numerous nodes. This requirement demands that the network interconnects have both low latency and high bandwidth. Currently, the most popular form of interconnect in high performance applications is InfiniBand (IB), because it offers both low latency and high bandwidth. The recently released 10 Gigabit Ethernet (10GigE) standard offers similar bandwidth and slightly worse latency performance as compared to the IB 4x. Despite its drawbacks, 10GigE can be more easily integrated into the common Ethernet network than IB because 10GigE uses the Ethernet protocol and IB does not. To determine 10GigE Message Passing Interface (MPI) performance, MVAPICH2 MPI benchmarks will be run over native InfiniBand and 10GigE with Remote Direct Memory Access (RDMA), and in addition MPICH2 MPI benchmarks will compare IP over IB(IPoIB) and 10GigE without RDMA. These benchmarks measure the MPI bandwidth and latency performance. The bandwidth comparison between 10 GigE with RDMA and InfiniBand are nearly equal but 10 GigE provides slightly higher bandwidth. With small message sizes the latency comparison shows IB with two to three times lower latency as compared to 10 GigE with RDMA. IPoIB and 10 GigE without RDMA both performed worse than IB and 10 GigE with RDMA, because the RDMA used by IB and 10 GigE allows for Operating System bypass(OS bypass). The poor test results of IPoIB and 10 GigE without RDMA show that OS bypass is key for high network performance. Even though the performance between IB and 10 GigE with RDMA are similar, IB still remains significantly cheaper than 10 GigE because 10 GigE is a newer technology. Future testing with production level MPI software should be conducted to determine what effect the latency difference between the two has on performance.

Processing Text Data and Advanced Search Capabilities for the Intelligence Community. JON DOBSON (East Tennessee State University, Johnson City, TN, 37614) LINE POUCHARD (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

One of the most valuable sources of intelligence for military agencies today is open source intelligence (OSINT). (1) OSINT embodies all publicly available information. It is becoming increasingly important to intelligence analysts as knowledge becomes globally available via the Internet. However, the problem with OSINT is that it is so time consuming and inefficient for an analyst to sort through all the available information on any given subject. The goal of this project was to use Semantic Web technologies and Natural Language Processing (NLP) to enhance the search capabilities of software tools used by intelligence analysts. The challenge behind this project was trying to avoid inheriting the search limitations that come with using only structured data or only natural language. In order to search structured data, the analyst must know exactly how the data is organized. Information Extraction based on automatic NLP techniques yields results that amount to 70% of results produced by a human. By combining the two methods with Semantic Web technologies like the Resource Description Framework (RDF), we are attempting to demonstrate that documents can be efficiently searched and cross referenced. OpenSource.gov files are automatically downloaded from an FTP server. After extracting the Extensible Markup Language (XML) files, all of the metadata is stored in a Java data structure. The text body is searched using Baseline Information Extraction (Balie) to find possible city names. For each possible city name, if it is found within a country listed in the XML metadata, then it is stored in the data structure. Finally, all of the data is used to create an RDF document, which is uploaded to a Sesame repository. Using Sesame’s web interface, it is possible to query, modify, and browse all of the data in the repository. A text file was also acquired from geonames.org containing information and coordinates on over 80,000 cities. Balie was chosen for named-entity recognition because it is easy to implement and highly configurable. An alternative to Balie, called General Architecture for Text Engineering (GATE) was also considered. However, since Balie can use GATE configuration files, Balie seemed like the most flexible choice. This project focuses on combining existing tools in a way that has not been done before. There is currently no standard way to convert XML data to RDF format, but this project is a step in that direction.

PTFrame: An Extensible Framework for Automating Parallel Performance Tools. MICHAEL SULLIVAN (George Mason University, Fairfax, VA, 22030) DR. WILLIAM GROPP (Argonne National Laboratory, Argonne, IL, 60439)

Second Order Boundary Derivatives in 3D. EMILY MITCHELL (University of South Carolina, Columbia, SC, 29225) LEONARD GRAY (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

The Boundary Element Method (BEM) is a numerical technique for the solution of the partial differential equations that describe physical phenomena. The integral equation formulation contains the Green's function (fundamental solution of the differential equation) and its first derivative; as these functions are singular, singular integral evaluation is an important aspect of the BEM. A boundary integral solution provides complete knowledge of the primary function (e.g., potential for the Laplace equation, displacement for elasticity) and its normal derivative on the boundary. An algorithm for the post-processing evaluation of second order derivatives of this primary function has been developed. The integral equation expression for these derivatives contains second (hypersingular) and third order (supersingular) derivatives of the Green's function, and thus the definition and evaluation of these highly singular integrals is the key task. In this work, the difference of the interior and exterior boundary limits of the integral equation is exploited to calculate the integrals. The difference procedure effectively weakens the singularities, resulting in finite limiting values, and moreover all non-singular terms vanish. The limit calculation is made possible by partial analytic evaluation of the four-dimensional parameter space integrals. The second derivative algorithm has been implemented for the 3D Laplace and Helmholtz equations. This latter case involves a more complicated Green's function, and it was necessary to use suitable Taylor expansions to carry out the limits. The ability to handle the Helmholtz Green's function indicates that the second derivative methods are generally applicable. The boundary element method is applicable to areas such as mechanics, acoustic and elastic wave propagation, and electromagnetism. This new second derivative algorithm should find application in many of these areas.

Software Development for a New OSCAR GUI that Will Support the Use of Virtual Machines. NATHANIEL GRAHAM (University of New Mexico, Albuquerque, NM, 87106) STEPHEN L. SCOTT (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Open Source Cluster Application Resources (OSCAR) facilitates the installation, administration, and programming of Beowulf type high performance computing (HPC) clusters. OSCAR has a variety of pre-packaged applications and utilities which allow one to install and manage clusters without specific technical expertise. An expanding interest in virtualization today has shown that virtual machines are a suitable solution to address issues for application, execution, and system management. However, the support of virtual environments in OSCAR is an ongoing effort, particularly regarding the user interface employed by OSCAR which is not currently optimized for management of virtual machines. The current OSCAR graphical user interface (GUI) was improved to better facilitate its use for managing virtual machines. The new OSCAR GUI was based on out of date OSCAR code. The project was divided into two phases: (i) the refactoring of existing code and (ii) the adaptation of the code to manage virtual environments. The base OSCAR code came from earlier GUI work that had fallen out of the main development path. That specific GUI project was never finished and was not maintained for a number of years. My first task was to modify the code to make it work with the current OSCAR architecture. It is now possible to launch the original GUI with all options available. The second phase was the adaptation of the code to virtual environments. For instance, compared to a standard execution platform, virtual environments are significantly more dynamic (virtual machines are created/destroyed on demand). For that I designed a new set of widgets for “node sets” and “package sets” management, and the mapping of virtual nodes onto physical nodes. The ultimate goal of my project was to provide a well-defined GUI for the management of virtual environments while maintaining the original Beowulf cluster installation abilities. This GUI will be integrated into OSCAR-V, the virtualized OSCAR infrastructure package, as the interface by which the user will manipulate OSCAR-V virtual environments.

Solving Ill-conditioned Linear Systems in High-Precision Arithmetic. JASMINE GARNER (Lane College, Jackson, TN, 38301) ED D'AZEVEDO (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

A linear system of equations is a set of linear equations in variables that can be represented in matrix form, where it is called the matrix coefficient of linear systems. A small change in the coefficients produces a large change in the solution and the system becomes ill-conditioned. The most efficient way for solving linear systems is the Gaussian elimination method, but this method itself is not always valid to the ill-conditioned cases. The goal of this project is to explore effective methods such as using high-extended, precision arithmetic library software for solving ill-conditioned block tridiagonal linear systems. The linear system is not too ill-conditioned but intermediate results may be ill-conditioned and can lead to instability. The idea is to use higher precision arithmetic to regain stability and accuracy. A dense direct solver using Gaussian elimination with partial pivoting from LAPACK (Linear Algebra Package) is modified to use higher precision. The solver computes a "LU" factorization where the matrix equals a lower triangular L multiplied by an upper triangular matrix U. The LAPACK is one commonly used library for performing numerical linear algebra computations. Both C++ and Fortran-90 high-level language interfaces are provided to permit one to convert an existing C++ or Fortran-90 program to use the library with only minor changes to the source code. LAPACK is written in Fortran since the language is commonly used for numerical computation. The code with extended precision was used to explore the range block sizes and number of blocks where the approach is effective by collecting runtimes and computed errors. A parameter (beta) controls scaling and conditioning of the system and as beta increases, the maximum error in solution increases as well. The measured runtimes decreases at the same rate as the number of block sizes increases in the data of fixed number of blocks. The results are summarized in plots. Although the runtimes are often lengthy, using a high-level program language to solve difficult linear equations permits faster development of large problems.

Someone is Spoofing My IP! Research and Development of a Detection Algorithm for IP Spoofing by Using Backscatter. DEBORAH KIMNACH (Illinois Institute of Technology, Chicago, IL, 60616) TAMARA MARTIN (Argonne National Laboratory, Argonne, IL, 60439)

Test and Evaluation of the Decision Evaluation in Complex Risk Networked Systems (DECERNS) Software. BRIAN TUCKER (SCCC, Selden, NY, 11784) TERRY SULLIVAN (Brookhaven National Laboratory, Upton, NY, 11973)

Environmental management requires decision support tools that are able to integrate multiple sources of technical information and evaluate options based on multiple criteria. Decision Evaluation in Complex Risk Networked Systems (DECERNS) is a web-based software system being developed with the objective of providing a methodology, computer models, and software tools to facilitate decision making in the field of sustainable land use. The system includes geographic information system (GIS) tools for mapping and data analysis along with risk models, economic analysis tools for evaluating costs, and tools for incorporating social choices into the decision process. The main advantage of DECERNS as opposed to existing decision support systems is its ability to integrate different models and measures as well as decision-maker values using Multi-Criteria Decision Analysis (MCDA) tools. The objective of this work is to perform extensive quality assurance testing of the system reliability, system functionality, and user interface to ensure the system can be deployed and used as desired. Testing was performed using Java version 1.6.01 under Microsoft Windows XP Professional. In addition, work included software development of Java applications for graphical display of MCDA results to rank and compare alternatives. The GIS map viewing function has been extensively tested and comments are being incorporated. At this time the full functionality of DECERNS has not been implemented. The next release will incorporate these comments and add the MCDA tools including the graphical display developed. This version will also undergo quality assurance testing as part of this program. Once the full functionality of this system is completed, it will be a valuable tool for environmental management.

The Integration of Sub-Hourly CALPUFF and its Processors. KATRINA LEE (Big Bend Community College, Moses Lake, WA, 98837) FREDERICK RUTZ (Pacific Northwest National Laboratory, Richland, WA, 99352)

CALifornia PUFF model (CALPUFF) is an Environmental Protection Agency (EPA) approved atmospheric modeling application used for the simulation of long range dispersion of airborne pollutants. CALPUFF is currently integrated into DUST TRANsport (DUSTRAN), a graphical dust dispersion modeling system. Within DUSTRAN, CALPUFF makes use of three preprocessor applications, a meteorological model, and a postprocessor application. The preprocessors condition raw topographical data for use in the models. The two models include a meteorological model named CALifornia METeorological model (CALMET) and the atmospheric dispersion model CALPUFF. Recent upgrades released for CALPUFF components provide new support for simulations on a sub-hourly basis. These upgrades necessitate changes to how data is acquired, processed, and output at all levels of the CALPUFF integration system. Integration of these new components is necessary to allow the system to function seamlessly as a single entity, automating the time consuming and human-error prone process of editing large amounts of input data while executing each processor in order. An integrated CALPUFF system may be used as a component of a larger system, such as DUSTRAN, with minimal input from the user. Integration of the new components is complete through the pre-processors and the CALMET model, with work still to be done on CALPUFF and the postprocessor application pending full documentation. This paper describes the progress, integration methods, and models for use in DUSTRAN.

Two Case Studies of Reversibility in Computational Methods. JOHN WRIGHT (Mercer University, Macon, GA, 31207) KALYAN S. PERUMALLA (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Currently state-saving is employed in many large simulations to realize rollback. Reverse computation is a recently proposed method which computes previous states instead of saving them. This approach can be beneficial on large machines as computing power is abundantly available and is possibly more efficient than retrieval from memory. This project investigates the reversibility of the well known Newton-Raphson root finding method and the possibility of developing a reversible interface for the Level 1 (vector) operations found in the Basic Linear Algebra Subprograms (BLAS) library. The mechanics of Newton-Raphson were studied and an algorithm was developed to reverse each iteration in the forward method. The reverse method involves finding the root of a derived function and forward Newton-Raphson is used on the derived function. Consequently, reverse Newton-Raphson gains the strengths and weaknesses found in Newton-Raphson. The reverse method has produced favorable results on functions that converge with forward Newton-Raphson. Unfortunately the forward method behaves unpredictably when multiple roots, periodic behavior, local minima, etc. exist in the target function and the reverse method will also produce unpredictable behavior in these cases. Further research is needed to handle unpredictability in certain functions in the reverse method. Routines in the BLAS Level 1 were analyzed and candidates were chosen based on the need for reversibility. Only those routines which modify input values require reversal. Reverse routines were developed for Givens rotation, vector scale, vector swap, and vector scale and update (saxpy). The reverse routines have shown identical scaling to their forward counterparts, however some problems concerning precision need to be resolved. Further work is needed to improve the interface to realize transparent reversibility for the vector copy operation. The forward BLAS library will need to be modified to implement copy reversibility.

Two-Dimensional Storage Analysis for Peta-Scale Computing. KYLE SCHOCHENMAIER (Iowa State University, Ames, IA, 50010) ROB ROSS (Argonne National Laboratory, Argonne, IL, 60439)

Using Kepler Scientific Workflows to Automate Processing of Data from Scientific Models. MATTHEW DEVRIES (Vanderbilt University, Nashville, TN, 37235) SCOTT A. KLASKY (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

As a Leadership Computing Facility, Oak Ridge National Laboratory (ORNL) is home to several large-scale scientific models that take advantage of the advanced supercomputing power available here. These models generate gigabytes of data, which must be archived, processed, and transferred to other locations before the output can be analyzed. Currently, each of these steps requires the use of separate tools, each with its own syntax and options. The aim of this project is to automate these processes in such a way that researchers can easily watch the results as a model is running. This is being done with Kepler scientific workflow software, which provides a flowchart-like interface for designing a script that can be executed from the command line. Using such a workflow that had previously been designed for a scientific model, new workflows are being developed for S3D, a state of the art direct numerical solver combustion model that runs on the entire Cray XT4 system. When finished, these will automate various parts of processing the data, as well as feeding into an AJAX dashboard, being developed separately, for real-time monitoring. For researchers using these models, this project should mean fewer tools to master and an easier time managing their data and monitoring their models as they run. On the development side, this project will provide feedback on the suitability of the Kepler software for this application, as well as possible improvements for the software and the process.

Using Math to Determine Optimal Placement for Wireless Access Points. KELLEN O'CONNOR (University of Tennessee, Knoxville, TN, 37919) DON WILLIAMS (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

Intranet and Internet networking have become a huge part of the operating strategies of many large organizations. Many organizations use networking to keep in contact with their employees, inform their employees, and to communicate among one another. It used to be that desktop computers were the standard, and laptop computers were only for the elite, used when absolutely necessary. Now, the world of electronics is focusing on smaller, more portable devices to make it more convenient for people to travel with their devices. So, the challenge for the organizations is to continue communication with these portable devices. The main method used is Wi-Fi, or wireless. Once a novelty, wireless communication has become a necessity in any large organization. Because wireless communication is a relatively new concept, there are not really any standards in place for the placement of wireless transmitters/receivers, wireless access points (AP’s). The goal of this project was to determine a mathematical equation that could be used to determine signal strength at different distances from the AP, so the AP’s could be placed in buildings by evaluating signal strength needs in certain areas. To determine this equation, I used a handheld signal strength measuring device. I setup the AP in a central location and recorded the signal strength at distances in intervals of five feet from the AP. I first measured the signal strength five feet from the AP, then again ten feet, fifteen feet, and so forth. I recorded that data and plotted the points on a graph. Visually, it was obvious to tell that the signal strength was linearly related to the distance from the access point. I used linear regression to determine an appropriate equation to relate the two variables, distance versus signal strength, and the data fit nicely. The results of this research are quite helpful in the information technology world; however, further research could also benefit the IT professionals. There are other factors to consider when setting up a wireless network that were not taken into account here. If research is continued, it would also be useful to obtain a device designed specifically for measuring signal strength. This would allow researchers to obtain more accurate data and therefore give better, more accurate results. This research is a step in the right direction and is the beginning of creating a useful tool for the IT professional when designing a wireless network.

Vascular Smooth Muscle Cell Migration Modeling. MATTHEW WOERNER (Tufts University, Medford, MA, 2155) JIM NUTARO (Oak Ridge National Laboratory, Oak Ridge, TN, 37831)

The migration of vascular smooth muscle cells is a catalyst for intimal hyperplasia, the thickening of the arterial walls in response to an injury in the blood vessel. Cell migration can be quantified using Boyden Chamber experiments. A model by Jabbarzadeh and Abrams model is used in designing this model. While this model is based on Jabbarzadeh and Abram’s model, our model contains several differences. In this algorithm, eukaryotic cells were modeled, instead of bacterial prokaryotic cells in their model, which slightly changed the nature of how cells move. A hybrid model, a model containing both discrete and continuous parts, was developed to simulate Boyden Chamber experiments in order to better understand cell migration. The hybrid nature of this model accounts for motion of individual cells, diffusion of the chemoattractant, and the interaction between the chemoattractant and cells. This model is being built using a C++ code and a discrete event system specification (DEVS). In the future this model will provide the foundation to aid in predicting the appearance of intimal hyperplasia following balloon angioplasty.

Wireless Network for On-Site Housing. JENNIFER SCHEUERER (Alfred State College, Alfred, NY, 14802) AJ TEMPROSA (Brookhaven National Laboratory, Upton, NY, 11973)

The implementation of a secure, functional, and reliable network is important when you are dealing with more than 2,000 users. Brookhaven National Laboratory (BNL) is in the process of upgrading the on-site apartments from Cisco 827 Routers to Cisco 857W Routers. The original network was designed to provide the scientists and visitors with network access for their accommodations. The original configuration of the Cisco 827 Routers was a Network Address Translation (NAT) on top of a Network Address Translation (NAT). This new configuration is now extending publicly routed space to the DSL router, where each apartment now has it’s own IP address. In this way it is simple to track down which apartment may be out of compliance with the Computer Usage Policy. On top of wired digital subscriber line (DSL), the new routers now provide each individual apartment with a unique wireless access point (WAP). The reason for the current architecture to be redesigned is to include additional services and capabilities as requested by user community and as required by the Department of Energy’s (DOE) security standards. The new design provides more functionality and security because now users are not connecting their personal wireless routers, which created many security holes in BNL network. The Cisco 857W’s were configured with a baseline and deployed within the project plan. The chosen encryption method for these routers is Wi-Fi Protected Access (WPA). This was chosen as the baseline configuration due to security flaws in the previously used Wireless Equivalent Privacy (WEP). All 112 routers were deployed with a baseline configuration on the Digital Subscriber Line Access Multiplexer (DSLAM) that met all the requirements of the new architecture. The redesign of the apartment network brought on multiple additional services and features for users and it also brought the apartment network to compliance with DOE’s security directives. This redesign also protects BNL from rogue wireless networks and it is now easier to quarantine machines and keep viruses from spreading onto BNL’s network. BNL is committed to providing all users with reliable and fully functional network access.

Computer Science Abstracts: