ITR-(ASE+NHS)-(sim+dmc): Non-Equilibrium Surface Growth and the Scalability of Parallel Discrete-Event Simulations for Large Asynchronous Systems

Rensselaer Polytechnic Institute, Troy NY

Investigators

Abstract

This award was made on a proposal submitted to the Division of Materials Research under the Information Technology Research solicitation NSF-04-012. It supports collaborative computational and theoretical research in the area of statistical physics. The PIs propose to develop algorithms for parallel discrete-event simulations, study their scalability and performance, and implement them on chosen applications involving materials related phenomena and statistical physics. The proposed research contributes to Advances in Science and Enginering (ASE) and to National and Homeland Security (NHS) as the ITR National Priorities. The proposed research focuses on innovations in computational modeling and simulations (sim) using innovative approaches to develop efficient communication and synchronization protocols in data intensive simulations (dmc) as Technical Focus Areas. The PIs will focus on parallel discrete event simulation algorithms which can be applied to study large-scale systems, including cell phone communication networks, models of spread of diseases (e.g., smallpox), the electric power grid, dynamic phenomena in materials systems, and ecological invasion in spatially extended environments. To understand the scalability and performance of large-scale massively parallel discrete-event simulations, the simulation itself is viewed as a complex interacting system consisting of tens of thousands of processors and an underlying network, facilitating communications and synchronizations between the processors. Using powerful tools and frameworks from statistical physics and particularly non-equilibrium surface growth, such as coarse-graining and finite-size scaling, the PIs identify the relevant node-to-node processes on the network. The universal features of the resulting non-equilibrium and stochastic model describe the progress of the individual processors in the parallel simulation (the parallel simulation "landscape"). Based on the "morphological" properties of this landscape, the PIs propose to design and develop algorithms that simultaneously optimize simulation speed and data management. The PIs plan to apply developed algorithms to various problems, including the study of dynamic phenomena in selected materials, and ecological invasion in multi-species models with preemptive competition. Both of these exhibit long-living metastable states with subtle finite-size effects. Implementation of massively parallel simulations is crucial to extract and to understand the underlying temporal and spatial patterns in these systems. This award also supports the education and training of students at the graduate and undergraduate levels in simulation and modeling with applications in science and engineering. The PIs endeavor to involve members of under-represented groups in research and education supported by this award. The research includes a collaborative component with researchers at the Los Alamos National Laboratory (LANL), where scalable simulation of individual-based models (e.g., for modeling a bioterrorist attack using smallpox in human-contact networks) and critical infrastructures (e.g., vulnerability detection for the electric power-grid) is of particular interest to the Laboratory. Graduate students will also participate in visits to LANL through the Laboratory's summer student program, further contributing to the education and training value of the proposed research. Research Experience for Undergraduates (REU) at Rensselaer will support, in part, undergraduate participation in the proposed research. The PIs will continue to participate in outreach activities, primarily through classroom interaction with science oriented high-school seniors. %%% This award was made on a proposal submitted to the Division of Materials Research under the Information Technology Research solicitation NSF-04-012. It supports collaborative computational and theoretical research in the area of statistical physics. The PIs propose to use powerful conceptual tools from statistical physics to develop algorithms to enable computer simulation of complex systems, including various materials related phenomena. The PIs main focus will be on simulations characterized by changes in the system that occur as discrete spatially localized events as time advances. This encompasses a diverse range of systems from, for example, battlefield simulation to the changes in the orientation of magnetic moments as a magnetic material is heated. The PIs will implement these algorithms and apply them to materials related phenomena and statistical physics. The proposed research contributes to Advances in Science and Enginering (ASE) and to National and Homeland Security (NHS) as the ITR National Priorities. The proposed research focuses on innovations in computational modeling and simulations (sim) using innovative approaches to develop efficient communication and synchronization protocols in data intensive simulations (dmc) as Technical Focus Areas. This award also supports the education and training of students at the graduate and undergraduate levels in simulation and modeling with applications in science and engineering. The PIs endeavor to involve members of under-represented groups in research and education supported by this award. The research includes a collaborative component with researchers at the Los Alamos National Laboratory (LANL), where scalable simulation of individual-based models (e.g., for modeling a bioterrorist attack using smallpox in human-contact networks) and critical infrastructures (e.g., vulnerability detection for the electric power-grid) is of particular interest to the Laboratory. Graduate students will also participate in visits to LANL through the Laboratory's summer student program, further contributing to the education and training value of the proposed research. Research Experience for Undergraduates (REU) at Rensselaer will support, in part, undergraduate participation in the proposed research. The PIs will continue to participate in outreach activities, primarily through classroom interaction with science oriented high-school seniors. ***