One emerging threat to the national critical infrastructure is the release of chemical or biological agents in or near large complex facilities. Predictive computational models for the time-dependent transport of gas and aerosol in and around target facilities is required to effectively assess and mitigate this threat. Despite this growing need for computational flow models that can be used in extremely large, complex geometries, current computational approaches are limited to 10 7-10 8 nodes. This is sufficient to model a single large interior space with a typical level of turbulence, but far less than the number of computation cells required to model all or even part of a large, modern building. The use of non-overlapping finite-element domain decomposition methods to achieve scalability with respect to problem size on massively parallel computers has been extensively developed over the past decade. In some cases, a high degree of scalability is achieved with only a single level of domain decomposition. Since a single level of domain decomposition is desirable for load balancing on massively parallel computers, a single-level method has the advantage of not introducing additional complex substructures on the problem, although special requirements may be imposed by the domain decomposition method on the nature of the subdomains to preserve scalability.
PDF
download
878987797
028-85220240
