High Performance Computing: Advancing USGS Science in the Era of Big Data
By Jeff T. Falgout, Janice Gordon, and Lisa Zolly
Figure 1. Decreased computation times for King model runs.
FL = Florida
OIC = ORNL Institutional Clusters
From processing highly detailed images of Mars to analyzing the genome of brook trout, USGS scientists explore complex, data-rich science questions that often require advanced computational power to produce critical simulations, models, and calculations. High performance computing (HPC) is the prevailing catalyst for answering today’s broad-scale earth science questions, providing maximum available processing capabilities to solve very large computational tasks. While local instances of HPC exist within the Bureau, USGS lacks a broadly accessible, national capability to serve the advanced computing needs of our research scientists.
A major strategic goal of the Bureau’s Core Science Systems (CSS) Mission Area is to “design and improve scientific computing systems, such as modeling frameworks, to allow computationally intensive operations to run through state-of-the-art technology platforms, from cloud-based resources to high performance computing centers” (Bristol and others, 2013). A recent conversation on the Bureau’s IdeasLab Forum, coupled with the rising number of USGS science projects requiring advanced scientific computing resources, spurred the CSS Core Science Analytics and Synthesis (CSAS) program to initiate a pilot project to assess the feasibility of a national capability for high performance computing and advanced computational capabilities to support USGS research. The project’s goals include:
- Evaluating the current state of computational capabilities within the Bureau
- Capitalizing on existing Interagency Agreements between the USGS and the Department of Energy’s Oak Ridge National Laboratory (ORNL), home to Titan, currently the world’s fastest supercomputer
- Assessing the value of USGS-owned computational clusters, including repurposed computer hardware made available through consolidation and termination of former USGS programs
- Developing a cooperative within the USGS Community for Data Integration that brings together HPC resource owners to exchange knowledge, coordinate purchases and software licenses, develop documentation, establish governance policies, and to advance HPC technology in the interest of the Bureau. Potential initial members include the Center for Integrated Data Analysis (CIDA), the Earthquake Science Center, the Astrogeology Science Center, the Wisconsin Water Science Center, the Center for Excellence in Geospatial Information Science (CEGIS), and the Florida Water Science Center; and
- Offering our researchers the ability to expand the scope of the science questions they are asking or possibly revealing new science questions, while significantly decreasing computation times for their projects
This CSAS pilot now supports several FY 2013 research projects requiring scientific computational resources and tools, from building Bayesian species models to modeling groundwater flow. One current participant is Jeffrey King, a research hydrologist at the USGS Florida Water Science Center, who runs parameter estimation and uncertainty analysis models to simulate transient groundwater flow in east-central Florida. These model runs traditionally take days, and even weeks, to run; using capabilities provided by the HPC pilot project, King has decreased his computation times by nearly 50 percent (Figure 1).
The HPC pilot project is providing critical context to the current and future computational needs of USGS scientists. Additionally, the pilot offers expertise, education, and assistance in navigating the sometimes complex pathways to external supercomputing facilities by leveraging partnerships with ORNL, the Department of the Interior, and academic institutions. Continued expansion of the pilot to include partnerships with existing local instances of USGS HPC capabilities will offer more USGS scientists access to the scientific computing resources they need to conduct broad-scale, computationally intensive science. In this era of “Big Data,” HPC capacity is a vital component in the acquisition and retention of world-class research scientists and their continued ability to provide reliable scientific information to describe and understand the Earth.
Kraken - A Cray XT5 supercomputer operated by the National Institute for Computational Science at the University of Tennessee and housed at ORNL under the Department of Energy.
Titan - A Cray XT7 supercomputer and currently the world’s fastest supercomputer is housed and operated by ORNL under the Department of Energy.
Bristol, R.S., Euliss, N.H., Jr., Booth, N.L., Burkardt, Nina, Diffendorfer, J.E., Gesch, D.B., McCallum, B.E., Miller, D.M., Morman, S.A., Poore, B.S., Signell, R.P., and Viger, R.J., 2013, U.S. Geological Survey core science systems strategy-Characterizing, synthesizing, and understanding the critical zone through a modular science framework: U.S. Geological Survey Circular 1383-B, 33 p.