PIs: Sanjay Govindjee and Greg Fenves
The Department of Civil and Environmental Engineering boasts over 40 outstanding faculty members and includes 718 graduate and undergraduate students. The department encompasses diverse research activities that span all engineering specializations associated with the maintenance, management, construction, analysis and design of civil infrastructure for human society. Primary areas of interest are construction engineering, environmental engineering, geotechnical engineering, structural engineering mechanics and materials, and transportation engineering. The Department is ranked among the top two civil engineering programs nationwide; the faculty are well-known in their specializations and include several Presidential Young Investigators, and members of the National Academy of Engineering, among many honors. Within the primary areas of research there are many on-going projects associated with large scale computations -- primarily physical system simulation such as solid mechanics, fluid mechanics, and transport mechanics. Below, we describe one of the major projects associated with Dr. Fenves' work on finite element methods (FEM) in solid mechanics that is expected to make substantial use of the equipment grant. The computing demands for civil engineering studies are very intensive whenever realistic problems are to be analyzed. This occurs because of the large physical dimensions treated and because the critical response of these systems, in many cases, relies on local detail. Often research is restricted to academic problems, which are small enough to understand the physics of the systems in only a limited sense. The analysis of real systems could greatly benefit from the availability of a network of parallel machines. It is widely recognized that to achieve the best possible performance on parallel and distributed computers, load balancing within an application is required. When implementing FEM on a parallel machine, static load balancing schemes have been typically employed. In this model the domain is partitioned into subdomains, one subdomain per processor, at the start of the application. This strategy, however, is not always sufficient because: (1) user load on the processors can change in time; (2) the initial partitioning might not be optimal; and (3) in nonlinear FEM analysis, certain regions of the mesh can become highly non-linear at any stage of the analysis, causing large changes in computational demand. Using a Network of Workstations (NOW) based on Intel processors, we will develop dynamic load balancing schemes for the subdomains to improve performance. The approach will be to modify the original partitioning by migrating elements between subdomains as the analysis proceeds. In previous work, we have tested different load balancing algorithmic heuristics, whose decisions are based on past and/or current and/or predicted performance measurements. The utilities will be implemented on the NOW using a recently developed object-oriented software environment for finite element analysis. The design of the framework allows for the seamless introduction of new elements, solution procedures, and load balancing algorithms into a FEM application. There is an increasing demand for the simulation of response under extreme loads such as wind and seismic movement. In current practice engineers have been restricted to small-scale FEM analyses that often neglect important effects. Much of this work, however, is currently being carried on individual PC-based workstations. The potential exists to greatly enhance the computational capabilities of the typical civil engineering firm without forcing them to abandon the platform base with which they are most comfortable. This project has great potential for making nonlinear finite element analysis a reality for the profession by use of relatively low cost networked workstations. The potential savings for society are tremendous -- both in terms of economic savings from lower cost structures to savings in human life from safer structures. Distributed processing capabilities using the Intel architecture will have a tremendous impact on the Department of Civil Engineering and its research. There are tremendous opportunities for collaborating with Computer Science and their various projects on parallel and distributed computing. A three year project is envisioned. The first year will concentrate on porting the object-oriented FEM framework to a group NOW. The second year will be devoted to experimenting with load balancing strategies. The third year will involve the scaling of problems to the campus NOW and building a library of parallel solution and load balancing utilities. Other specific research programs that are expected to use and benefit from this equipment grant include: (1) modeling of composite microstructures by Monte-Carlo methods (Govindjee); (2) localization modeling in continuum and structural theories (Armero); (3) multigrid equation solving for finite element systems (Taylor); (4) simulation studies of micro-machines in bio-medical research (Ferrari); (5) large structural system dynamic response analysis for seismic loading (Fenves); (6) coupled thermo- and diffuso-mechanical finite element analysis (Govindjee, Armero); (7) contaminant transport modeling in soils and ground water (Sitar, Rubin); (8) stochastic modeling of site response and liquefaction (Sitar, DerKiureghian, Seed); (9) discontinuous deformation modeling of granular materials (Bray, Sitar); and (10) atmospheric transport modeling of air pollution (Harley).
February 1999