CAAR
SCEC is preparing a Center for Accelerated Application Readiness Proposal (CAAR) to participate in OLCF development of their next generation supercomputer called Summit.
Title
Physics-based Urban Seismic Hazard Model Application Readiness
PI: Thomas H. Jordan [1]
Collaborators: Jacobo Bielak [2], Scott Callaghan [1], Yifeng Cui [3], Kim Bak Olsen [4], Philip J. Maechling [1], and Ricardo Taborda [5]
[1] Southern California Earthquake Center/University of Southern California, [2] Carnegie Mellon University, [3] San Diego Supercomputer Center, [4] San Diego State University, [5] University of Memphis
Abstract
Southern California Earthquake Center (SCEC) geoscientists and computer scientists propose to collaborate with researchers at Oak Ridge Leadership Computing Facility (OLCF), other Department of Energy (DOE) computing centers, and DOE technical collaborators, on a Center for Accelerated Application Readiness (CAAR) project to establish application readiness for SCEC’s physics-based seismic hazard modeling software. The physics-based computational software developed and optimized by this collaboration will calculate physics-based predictions of ground motions that incorporate realistic fault rupture models and 3D geological structures and run efficiently on the next generation of DOE supercomputing systems. By the end of the CAAR program, SCEC will use DOE leadership computing resources, together with the CAAR optimized computational software, to calculate physics-based urban seismic hazard models for cities in the United States.
SCEC’s application readiness software development will focus on existing high-performance earthquake ground motion simulation codes that include and form the basis for SCEC’s CyberShake physics-based seismic hazard computational platform. In particular, the proposed SCEC/CAAR application readiness collaboration will focus on developing and extending the efficiency, scalability, and portability of three core applications: (1) AWP-ODC, a finite difference dynamic rupture and earthquake wave propagation code with both CPU and GPU implementations, (2) Hercules, a finite element earthquake wave propagation code with both CPU and GPU implementations, and (3) CyberShake, a workflow-oriented, heterogeneous, seismic hazard computational platform that performs probabilistic seismic hazard calculations by running ensembles of earthquake simulations using AWP-ODC or Hercules, and then hundreds of millions of loosely-coupled post-processing calculations. Computational improvements to AWP-ODC and Hercules will help achieve the regional scale, high-resolution, high-frequency earthquake simulations required for CyberShake to maximize the value of the computational results for use by the engineering community, who use seismic hazard maps to improve the design of infrastructure to withstand strong ground shaking. Computational improvements to CyberShake will enable the calculation of improved probabilistic seismic hazard models at 1.5 Hz and higher, and will help DOE confirm that their emerging extreme scale computing environments support research calculations requiring ensemble and automated repeatable multi-stage research calculations.
We believe it is important to improve the scalability of these three codes as a group. Two of the codes, AWP-ODC and Hercules, provide alternative methods for solving equivalent deterministic earthquake wave propagation problems, and it is important to maintain alternative solution methods as we increase the problem scale, complexity, and resolution. For critical engineering simulation applications, such as seismic design limits of mid- and high-rise buildings or other infrastructure systems with a wide range of stiffness/flexibility (e.g., roads and bridges, power generation and transmission facilities), it is important to show equivalent results using alternative methods, as a way to confirm simulation results. Increasing the scale of the CyberShake workflows is important to ensure it is possible to execute end-to-end ensemble calculations, using the earthquake wave propagation software. The CyberShake platform produces well established and widely used computational data products needed by the broad impact users of urban seismic hazard models.
