University of Central Florida Research Participants

  • Civil Engineering
  • Institute for Simulation and Training
  • Department of Physics
  • Mathematics
  • Nanoscience Technology Center

Research Topics

The University of Central Florida has procured a high performance computer capable of supporting research in several areas. The areas and type of research include the following;

Real time interactivity

Typically High Performance Computing, characterized by 64 bit word length, MPI, high speed interconnects, large amounts of local and spinning memory, and appropriate operating system features (e.g., load balancing) has typically been reserved for batch processing. Also, HPC machines are typically procured for a specific class of problem that might need large amounts of cache, other memory, CPU cycles, or inter-processor communications. Interactive computing brings a potential new challenge with respect to minimizing end-to-end system latency and allowing runtime parameter change where end-to-end implies an input from a user and output to a user in real time (say 30Hz) at less than a 100 ms latency and parameter change implies changing input variables during runtime. The principal type of interactions we are interested in involve large numbers of computer generated entities that represent people and vehicles which move over simulated terrain and buildings and interact with the environment and each other. These entities are represented in a physically realistic manner, have realistically represented behaviors appropriate to the scenario being depicted, and respond appropriately to human users. Additionally, these entities operate both individually and in groups. In both individual and group cases it is of interest that direction given to the multiple entities be transmitted and acted upon in a timely manner. Given the feasibility of this approach, extensions include several scenarios operating on the same environment, but autonomously from each other. Software for these applications exists in the gaming and military communities, but is generally limited to a few hundred entities operating under the auspices of a controller.
HPC also facilitates scalability of computational models implemented in medical, physical science, and engineering disciplines via the implementation of parallelized algorithms and commercially available software packages. Below are listed some applications commonly implemented in these disciples which benefit from HPC.

Electromagnetic Finite Difference Modeling Work

IST researchers are developing electromagnetic simulation models for measuring Electromagnetic (EM) wave propagation and interaction with different objects. The project aims at studying Radar signatures and target recognition algorithms to determine efficient methods for real-time target recognition and classification. Currently they are using a free-source software simulation package called MEEP. MEEP is supported and portable to any Linux/Unix based O/S system. Their current application executes on a single processor machine that calculates the time stepped field potentials as the EM wave propagated through the computational domain. As the frequency of the wave is increased in the simulation, the wave length correspondingly decreases requiring a finer mesh generation for the Finite difference models. This leads to higher computational and memory requirements.
The MEEP software can be used on the STOKES machine as the parallel version of the MEEP software uses MPI and distributed memory for processing large problems. Using MEEP in a parallel environment allows us to study large scale models such as aircraft and ships at an improved resolution than possible on a single processor machine. The current MEEP application is being run on a distributed memory based OPCODE cluster front-end and two nodes with 1 GB memory on each node, 20 GB hard-drives and 1.4 GHz AMD Thunderbird processors.

Data Mining Applications for Health Applications

The process of extracting useful information from a set of data values is called “data mining”. This data can be used to create models to make predictions. The College of Health and Public Affairs is interested in using STOKES to run adapt mining applications and programs such as DTREG, a software for Predictive Modeling and Forecasting to analyze patient health records and determine vital statistics such as drug interactions and possible side-effects the patients might undergo when taking drugs for different diseases. They also use SPSS, SAS and GIS software applications for studying geographical patterns and statistics for patient records. One of the primary requirements for their applications is very high processing speed capability, large memory capacity and any version Linux O/S. Their current applications are being hosted on 4 Sun Microsystems Netra 440 servers and 220 servers with 1 TB of Dell PowerEdge Storage. But will soon be running on STOKES.

Physics and Nano-Technology

Nanotechnology has emerged as a new field where design and engineering of systems such as hydrogen fuel cells or solid oxide fuel cells require consideration of electronic structure configuration at the atomic level. Realistic modeling of nano-system behavior and function, as well as modeling self-assembly processes and growth, requires molecular dynamics and quantum mechanical electronic structure models of hundreds to thousands of atoms over millions of iterations. Parallel molecular simulations that calculate empirical force fields naturally lend themselves to linear scaling and benefit from STOKES with hundreds of processor cores in the LINUX computational environment.

Weather Research & Forecasting (WRF) Model

IST researchers are proposing to build and run the Weather Research & Forecasting (WRF) Model on high-performance computing clusters at the University of Central Florida. This project aims at promoting the development of climate and weather models as well as research into real-time forecasting and idealized weather/climate research. Utilizing STOKES as a viable HPC is beneficial for the IST researchers.
The Weather Research and Forecasting (WRF) Model is a next-generation mesoscale (regional – 1 to 100 km) numerical weather prediction system designed for both operational weather forecasting and atmospheric research. The WRF system architecture is modular with multiple dynamical cores, a 3-dimensional variational (3DVAR) data assimilation system allowing for computational parallelism and system extensibility. WRF is suitable for a broad spectrum of applications across scales ranging from meters to thousands of kilometers.
WRF is portable to many platforms including UNIX, Linux, HPUX, AIX, IRIX, Solaris, UNICOS, MacOS, ALPHA Tru64, and Windows CCS. It is also very scalable. Although it is typically run in multi-processor (parallel) mode with distributed memory, it can be run serially on a single processor and will scale to massive parallel computation on HPC and high availability (HA) clusters with shared and distributed memory and storage.

Modeling of Hurricane Storm Tide Induced Bridge Pier Scour for the Florida Coasts

The estimation of tropical-cyclone-generated waves and surge (i.e., hurricane storm tides) in coastal waters and the near-shore zone is of critical importance to the assessment of potential damage to coastal infrastructure (namely bridge piers) in the event that a storm makes landfall. While significant progress has been made in the theoretical ability to deterministically model bridge scour, bridge scour modeling in coastal areas is constrained due to computational limitations. These computational constraints are overcome by implementation of STOKES.
If the intense flow of water from a hurricane storm tide results in the undermining of bridge piers (commonly referred to as bridge pier scour or bridge scour), then the failure of the bridge can result in on the order of billions of dollars of economic damage to the public. The cost would be associated with significant bridge repair and/or reconstruction as well as significant hampering of the local and state economy. Alternatively, a direct financial benefit on the order of tens to hundreds of millions of dollars can result if state-of-the-art bridge pier scour modeling can supply models that will affect how the intense flow of water from a hurricane storm tide results in the undermining of bridge piers (commonly referred to as bridge pier scour or bridge scour), then the failure of the bridge can result in on the order of billions of dollars of economic damage to the public. The cost would be associated with significant bridge repair and/or reconstruction as well as significant hampering of the local and state economy. Alternatively, a direct financial benefit can result if state-of-the-art bridge pier scour models can show that coastal bridges may be designed so that future construction costs may be reduced.

Crowd Modeling

The type of problem which the crowd dynamics group is currently simulating is basically a multi particle system. It is very comparable to simulations run on solid state dynamics. Since each particle is only affected by a small group of individuals near it, and that each calculation is done independently of all other calculations, the computations are well suited for parallelization. The data stored for each particle is relatively small so there are not large memory constraints; most of the bottleneck is the actual computations. Relatively large data files are generated, so the amount of available disk space is also important.
Currently IST is doing small scale simulations on PCS and having them run over weeks to generate the needed datasets. The implementation is currently in JAVA and has been developed to run on a Linux platform. There are a few other data analysis tools that we have implemented which are also constructed to run in a Linux or windows environment. Currently simulations of around 200 individuals require 12 hours on a single processor with 1GB memory. We envisage expanding to simulations of 20,000 individuals (UCF Brighthouse stadium scale) so that the equivalent of several hundred processors could be utilized.

Graphics Work

An important component of high performance computing is visualization. To meet this need, the IST cluster will require multiple graphics adapters to create high resolution tiled display environments. The hardware should be the latest COTS technology to take advantage of GPU advances (like pixel shader programs) and hardware controlled genlocking. Furthermore, to facilitate rendering of the information from the cluster computation nodes, a large shared memory space will be required for optimal access by the graphics processors.
The visualization uses would include high-definition rendering of a variety of problem data and enable interactive rendering. Cluster users would have the possibility of changing experimental parameters and see the visualization change accordingly. Moreover, by using novel interface device types, the effects of collaborative interaction by multiple users can be visualized within the experimental data.

Hardware Requirements

Researchers implementing severe climate models, hydrogen fuel cell models, nanoscale optical response models, and medical data mining techniques utilizing open source, commercially available, and licensed applications and libraries mentioned above require a system that maximizes a union between compatibility an performance. It has been the experience of researchers in various fields that the importance of system resources in a high performance computational system should be listed as follows: processor type and clock rate, amount and speed of memory available, storage, and interconnect. STOKES satisfies all these requirements, as seen on the Hardware page.