Source: UTAH STATE UNIVERSITY submitted to NRP
HIGH PERFORMANCE COMPUTING EDUCATION AND RESEARCH FOR AGRICULTURAL APPLICATIONS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
Reporting Frequency
Annual
Accession No.
0214051
Grant No.
2008-34610-19175
Cumulative Award Amt.
(N/A)
Proposal No.
2008-03714
Multistate No.
(N/A)
Project Start Date
Jul 1, 2008
Project End Date
May 31, 2010
Grant Year
2008
Program Code
[CC-F]- High Performance Computing, UT
Recipient Organization
UTAH STATE UNIVERSITY
(N/A)
LOGAN,UT 84322
Performing Department
AGRICULTURAL EXPERIMENT STATION
Non Technical Summary
The project goal is to extend the use and application of high performance computing to the agricultural research community. The Center for High Performance Computing at Utah State University (HPC@USU) addresses this goal by developing an integrated, easy to use computational infrastructure. This initiative is in concert with the President's Information Technology Advisory Committee (PITAC) report and the recently passed High Performance Computing (HPC) Revitalization Act and addresses the critical needs for high performance computing education and resources on the local, state and national level. Through this program, the high performance computing resources at HPC@USU will be expanded by storage and computing capabilities. The state and national needs for high performance computing education, in this case for agricultural scientists and students, will be addressed by the development and dissemination of training material specifically targeting the agricultural research audience. We will demonstrate the value of high-performance computing through the following projects related to agriculture: 1. Seed grants to develop new innovative projects using advanced computational technology. 2. High resolution Utah snow pack simulations. 3. Pore scale pathogen transport simulations in soil. 4. Sediment transport in small streams. 5. Benchmarking of the open source simulation codes for multi core processors. 6. Implementation of parallel I/O for the simulation codes used in above projects.
Animal Health Component
(N/A)
Research Effort Categories
Basic
100%
Applied
(N/A)
Developmental
(N/A)
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
1024010208025%
1120210208025%
9037299208050%
Goals / Objectives
The project goal is to extend the use and application of high performance computing to the agricultural research community. The Center for High Performance Computing at Utah State University (HPC@USU) will address this goal by developing an integrated, easy to use computational infrastructure to address needs of the agricultural community accessible to all faculty, students and potentially USDA researchers. We will provide seed grants to researchers in agriculture related areas to create new innovative research projects using high performance computing technology. These research grant will provide funding for interdisciplinary projects. To moderate snow trends and variability, reservoirs and canals have been built to meet growing needs of agriculture in Utah where the population has almost doubled over last three decades. Nevertheless, an adequate knowledge of snow as the primary water source is still needed for successful water management. In this mountainous area with extremely varied topography, without densely and evenly distributed observing stations, numerical computer models provide unique tools to understand historical hydroclimate processes and phenomena and predict these processes and phenomena for the future. We plan to use a state-of-the-art high-resolution regional climate model to simulate and predict the detailed spatio-temporal distribution of snow pack in Utah. We will seek to quantitatively characterize the long-term trends and year-to-year variability of snow pack and associated temperature and precipitation changes over Utah, to more accurately simulate the warming-induced spatial shifts of these trends and variability, and to predict changes in snow and related climate variables for the mid- and late 21st century under a moderate carbon increase scenario. We will also develop an open source simulation program for modeling bacteria movement in soil. A detachment model will be integrated in the existing code. This will then be used to simulate cryptosporidium oocyst transport on a pore scale. Next, we will implement a model which characterizes the transport of motile bacteria through porous media using two effective transport parameters, random motility and chemotactic sensitivity. We will also perform Large Eddy Simulation (LES) for stream sedimentation by integrating LES models in an existing volume of fluids solver. All the above simulation codes need to run on modern computing platforms, therefore, we will characterize the performance of these simulation programs on modern multi-core processors. We will also implement parallel I/O in the pore-scale and stream simulation codes. Educational material and examples on how to use this cyberinfrastructure will be created and disseminated through the web. We will organize a symposium on high performance computing in agricultural research. Speakers from academia, government and industry will be invited to present about the importance and their successful use of high performance computing in agriculture.
Project Methods
In this mountainous area with extremely varied topography, without densely and evenly distributed observing stations, numerical computer models provide an unique approach to understand historical hydroclimate processes and phenomena and predict these processes and phenomena for the future. We plan to use a state-of-the-art high-resolution regional climate model to simulate and predict the detailed spatio-temporal distribution of snow pack in Utah. We will seek to quantitatively characterize the long-term trends and year-to-year variability of snow pack and associated temperature and precipitation changes over Utah, to more accurately simulate the warming-induced spatial shifts of these trends and variability, and to predict changes in snow and related climate variables for the mid- and late 21st century under a moderate carbon increase scenario. This much-needed study is only possible when high performance computing facilities are available. These facilities include massive parallel computing nodes, high capacity and performance hardware storage, and high-end visualization equipment. However, the current high-end computing resources at the Utah State University are limited to perform such a study, and thus, the expansion of these resources through this project are necessary. For the transport of in soil we will use a computational fluid dynamics code with integrated and coupled particle transport. We will implement a detachment and a random motility and chemotactic sensitivity models for the pathogen simulations. For the stream simulations the Large Eddy simulation approach will be used. This approach coupled with the volume of fluid method provides the basic solver for unsteady free surface flow. This solver will then be coupled with a particle transport approach and blockage based on a porosity approach. Benchmarking using hardware performance counters will be applied to characterize the performance of the simulation codes described above. This will allow us to select the right hardware platform for these simulation codes. Additionally, since all of these simulations will generate terabytes of data, parallel I/O will be implemented based on MPI and HDF.

Progress 07/01/08 to 05/31/10

Outputs
OUTPUTS: Engineering graduate students at USU are engaged in this multi-disciplinary project designed to develop a full-scale computational fluid dynamics (CFD) model of a river with sediment transport. We collaborate with researchers from the USU Natural Resources department to obtain a realistic river topography, which we use to create a computational mesh with adequate resolution for Large-Eddy Simulation (LES). This process involves four major steps: 1. Design a computational mesh from field data; 2. Select an appropriate turbulence model; 3. Create a particle tracking algorithm to simulate the sediment transport; 4. Incorporate all parts into one overall computational simulation. We will make the developed solvers available after the validation phase. In our work we evaluated the mobilization and behavior of fine-grained quartz particles with silica grains, from a numerical simulation standpoint. The colloidal and shear forces were computed in accordance with established methods and validated with available literature. These, along with the drag and gravitational forces, were made manifest through their contributions to the momentum source terms associated with the Navier Stokes equations of motion. The use of OpenFOAM, and in particular, the Ico-LagrangianFoam solver, greatly facilitated the creation of the numerical algorithm. The results were validated against the experimental, visualization experiments of Cerda, and showed that the response of the fines particle, at close separation distances, was largely a function of the chemistry of the fluid medium. That is, for moderate to low levels of shear force, high values of pH and low values of electrolyte concentration constitute a repulsive response. Whereas, for low values of pH and high levels of electrolyte concentration, the attractive Van der Waals force dominates, and deposition is witnessed. We have tested the Weather Research Forecasting Model (WRF), coupled with the Community Land Model (CLM) version 3.5, on the USU high performance-computing cluster. The results indicate that the snow simulations are dramatically improved with this coupled model. Further analysis indicates that such improvement is the result of more realistic allocation of surface energy in CLM when compared to two land surface schemes embedded in the previous version of WRF. In addition, the improved snow simulation further reduces overestimated precipitation and a warm bias in WRF. Additional simulations indicate that topography plays an important role in snow modeling. CLM at 10 km resolution produces the most accurate snowpack simulations when compared to the simulations at coarser resolutions where the snowpack is underestimated. Moreover, we can see that CLM still can produce realistic snowpack simulations at 20 km resolution when the prescribed elevations in the model are replaced with the observed values, further indicating the important role of topography in snowpack simulations. PARTICIPANTS: Overall project coordinator, Jiming Jin, PI, snowpack simulations, Nate Benson, Co-PI, system manager, E. Garbi, M. Hradisky, P. Villanueva, A. Zabriskie - sediment transport simulations, K. Horne, J. McCulley - CGNS parallel I/O implementation, T. Johansen, S. Ripplinger - OpenFOAM, W. Frisby, T. Quist - system help; Lijuan Wen - Regional climate modeling. TARGET AUDIENCES: The findings from this project will give state water managers with informed knowledge about water availability in the western United States. PROJECT MODIFICATIONS: In addition to snow simulations over the western United States, we performed irrigation simulations over China on the USU HPC Cluster.

Impacts
The project was well-received and showed how computational fluid dynamic (CFD) tools can take advantage of parallel storage solutions through a parallel implementation of the CFD general notation system (CGNS). The current CGNS system provides a standardized and robust data format to the CFD community and has made it easier to exchange information between different tools in the CFD process. However, it lacks an implementation to take advantage of fast parallel storage systems. The HPC@USU project solves that problem and enables CFD developers to take advantage of parallel systems with little additional parallel programming effort. The team was encouraged by the panel of judges to continue this work because a large number of HPC users will be able to benefit from this library. The significant improvements of snow simulations in our regional climate model will greatly benefit agricultural water use forecasts in the western United States. The improved version of our regional climate model will give the climate community a better tool for climate research and forecasts. Through these presentations, the tools of HPC@USU have been advanced and disseminated to a much wider audience in the agricultural and life sciences fields, including the USU Colleges of Agriculture, Engineering, Science, Natural Resources, Water Laboratory, as well as SC09 conference attendees and USU Spring Runoff attendees through Utah and nation-wide.

Publications

  • P. Wu, J. Jin, & X Zhao. 2010. Impact of climate change and irrigation technology advancement on agricultural water use in China. Climatic Change (in press).
  • K. Horne, N. Benson, and T. Hauser. 2009. An Efficient and Flexible Parallel I/O Implementation for the CFD General Notation System, SC09 Storage Challenge finalist presentation. Hauser, T. 2009. Benchmarking serial and parallel CGNS I/O performance, in '47th AIAA Aerospace Sciences Meeting including The New Horizons Forum and Aerospace Exposition.'
  • Hauser, T. 2008. Benchmarking the CGNS I/O Performance, in '46th AIAA Aerospace Sciences Meeting and Exhibit.'
  • Hauser, T. & Allen, J. 2007. Numerical simulation of the behavior and mobilization of fine-grained quartz particles in porous media, in 'OpenFOAM International Conference.'
  • Hauser, T., Allen, J. & Ripplinger, S. 2007. Numerical Simulation of the behavior and mobilization of fine-grained quartz particles in porous media, in '60th Annual Meeting of the Division of Fluid Dynamics.'


Progress 07/01/08 to 06/30/09

Outputs
OUTPUTS: For this project, we evaluated the mobilization and behavior of fine-grained quartz particles with silica grains, from a numerical simulation standpoint. The colloidal and shear forces were computed in accordance with established methods and validated with available literature. These, along with the drag and gravitational forces, were made manifest through their contributions to the momentum source terms associated with the Navier Stokes equations of motion. The use of OpenFOAM, and in particular, the Ico-LagrangianFoam solver, greatly facilitated the creation of the numerical algorithm. The results were validated against the experimental, visualization experiments of Cerda, and showed that the response of the fines particle, at close separation distances, was largely a function of the chemistry of the fluid medium. That is, for moderate to low levels of shear force, high values of pH and low values of electrolyte concentration constitute a repulsive response. Whereas, for low values of pH and high levels of electrolyte concentration, the attractive Van der Waals force dominates, and deposition is witnessed. One important, often overlooked, issue for large, three dimensional time-dependent computational simulations is the input and output performance of the solver, especially for large time-dependent simulations. The development of the CFD General Notation System (CGNS) has brought a standardized and robust data format to the CFD community, enabling the exchange of information between the various stages of numerical simulations. Application of this standard data format to large parallel simulations is hindered by the reliance of most applications on the CGNS Mid-Level Library. This library has only supported serialized I/O. By moving to HDF5 as the recommended low-level data storage format, the CGNS standards committee has created the opportunity to support parallel I/O. We developed a parallel implementation of the CGNS Mid-Level Library and an I/O request-queuing approach to overcome some limitations of HDF5. We have tested the Weather Research Forecasting Model (WRF), coupled with the Community Land Model (CLM) version 3.5, on the USU high performance-computing cluster. The results indicate that the snow simulations are dramatically improved with this coupled model. Further analysis indicates that such improvement is the result of more realistic allocation of surface energy in CLM when compared to two land surface schemes embedded in the previous version of WRF. In addition, the improved snow simulation further reduces overestimated precipitation and a warm bias in WRF. Additional simulations indicate that topography plays an important role in snow modeling. CLM at 10 km resolution produces the most accurate snowpack simulations when compared to the simulations at coarser resolutions where the snowpack is underestimated. Moreover, we can see that CLM still can produce realistic snowpack simulations at 20 km resolution when the prescribed elevations in the model are replaced with the observed values, further indicating the important role of topography in snowpack simulations. PARTICIPANTS: Thomas Hauser, Jiming Jin, snowpack simulations, E. Garbi, M. Hradisky, P. Villanueva, A. Zabriskie - sediment transport simulations, K. Horne, J. McCulley - CGNS parallel I/O implementation, T. Johansen, S. Ripplinger - OpenFOAM, W. Frisby, T. Quist - system help; Lijuan Wen - Regional climate modeling. TARGET AUDIENCES: agricultural research community climate community PROJECT MODIFICATIONS: A new cluster computer was purchased with 64 nodes containing dual quad core AMD processors and a fast infiniband interconnect. This will better enable the proposed climate and fluid simulations, since our initial benchmarks reveal that we need much more computing power than initially estimated. We used the seed grant and symposium budget as approved by USDA to purchase the cluster. The cluster is now operational.

Impacts
As a result of our work, my student and I were selected as one of four finalists in the prestigious international Supercomputing Conference SC09 Storage Challenge. The project shows how computational fluid dynamic (CFD) tools can take advantage of parallel storage solutions through a parallel implementation of the CFD general notation system (CGNS). The current CGNS system provides a standardized and robust data format to the CFD community and has made it easier to exchange information between different tools in the CFD process. However, it lacks an implementation to take advantage of fast parallel storage systems. The HPC@USU project solves that problem and enables CFD developers to take advantage of parallel systems with little additional parallel programming effort. We presented our work at the November SC09 conference that has an acceptance rate of less than 20%. Our entry didn't win, but we were encouraged by the panel of judges to continue this work because a large number of HPC users will be able to benefit from our library.

Publications

  • No publications reported this period