Project Title/ID Number NEES Computational Simulation - High End Computing—4222003
Start/End Dates 10/1/03—3/31/04
Project Leader Gregory Fenves (UCB/F)
Team Members Filip C. Filippou (UCB/F)

F=faculty; GS=graduate student; US=undergraduate student; PD=post-doc; I=industrial collaborator; O=other

Click on images to enlarge in a new window

1. Project Goals/Objectives:

The goal of this project is to port, test, and apply OpenSees (Open System for Earthquake Engineering Simulation) on parallel computing resources. Parallel computers available at supercomputer centers, such as NCSA and SDSC are emphasized, so that OpenSees can be used as a NEESgrid application resource in the George E. Brown, Jr., Network for Earthquake Engineering Simulation.

2. Role of this project in supporting PEER’s mission (vision):

PEER’s research mission to develop performance-based earthquake engineering methods is highly dependent on improved capability for simulation the response and performance of structural and geotechnical systems. As simulation models have increased in size and complexity, particularly during the testbed projects, it has become clear that large-scale simulations require more computing power. This project will address the need for conducting large-scale simulations of structural and geotechnical systems on high-end computers.

3. Methodology Employed:

OpenSees software has been designed for parallel computers in a number of ways. The first is that data structures (e.g. for elements, nodes, etc.) can be communicated between different memory spaces on different computers. Secondly, analysis procedures can be associated with a method for numbering DOF using graph partitioners (e.g. Metis, Zoltan). These capabilities support domain decomposition methods, and can be used to implement static and dynamic partitioning. Analysis procedures can be used for computing the Shur complement (that is static condensation) necessary for domain decomposition. Additionally, the interface with solvers is general enough to be able to interface with parallel solvers (such as Parallel SuperLU). We have generally found that iterative solvers are not robust enough for highly nonlinear problems. Although implicit time stepping procedures have been used primarily to-date, the OpenSees software architecture is designed to support explicit methods, which can also be efficiently parallelized.

4. Brief Description of past year’s accomplishments (Year 6) & more detail on expected Year 7 accomplishments:

This is a new project this year. By the end of Year 7, we expect to have completed the following: porting and testing of OpenSees to SDSC computers (which will be the home of the NEES NITSC), and possibly other resources, time and access permitting; conducting scalability studies on large simulation from PEER research; examining the performance of Parallel SuperLU, pending the UC Berkeley group completing the API for the parallel solver; documentation and posting of software and results on the OpenSees website to provide guidance for users.

5. Other Similar Work Being Conducted Within and Outside PEER and How This Project Differs:

Within PEER, Jeremic (project 422) is developing geotechnical models and investigating parallel computing applications in coordination with this project. The Humboldt Bay bridge test bed project (Conte, project 403) will be the principal example we will investigate for scalability and speed-up because of the large 3D model used in the testbed.

Outside of PEER there is relatively little ongoing work on parallel applications for earthquake engineering simulation, although there is a vast work amount of work in computational mechanics, seismology, and mathematics (for solvers). We have utilized much of this research in the software design for OpenSees to allow implementation of the methodologies or interfaces with the solver software.

6. Plans for Year 8 if project is expected to be continued:

This is a one-year project only.

7. Describe any actual instances where you are aware your results have been used in industry:

None to date.

8. Expected Milestones & Deliverables:

We expect the port to be completed in June 30, 2004, which will coordinate with the NEESgrid release. Scalability studies will be completed by September 1, 2004, and the documentation and posting completed by September 30, 2004.

back to top