Sponsor: NSF
This study aims to explore a range of computational issues related to large scale simulation of n-body systems. It investigates relationships between applications, algorithms, and architecture needs to identify favorable operating points. As potential estimation becomes faster, other components of the system present serious bottlenecks. These include data analysis, interpretation, and visualization. Effective algorithms for pre-processing this data as it is generated along with fast techniques for querying pre-processed data are necessary. These issues form the focus of this project. The contributions of this project are: (i) Analytical quantification of performance, errors, and parallel processing overheads; (ii) Methods for dynamically steering computation to sub-domains that exhibit "interesting" properties such as fractures, dislocations, and defects; (iii) Techniques for predicting macro-properties from molecular dynamics simulations; and (iv) Methods based on selective focus and level-of-detail for data analysis and visualization.