Application-Specific Computing Structures for Large-Scale Molecular Dynamics Simulations.

Molecular dynamics is a computer simulation method used in the study of condensed matter, particularly liquids and phase changes. This is done by integrating the equations of motion of a number of representative particles over a specified period of time, studying their movements, and keeping track of the system's physical properties as it evolves. The technique has grown in popularity, and larger more complex simulations are continually being proposed, but current supercomputers cannot solve these problems in reasonable computation times. The main objective of this research was to design a computing structure optimized for molecular dynamics problems. The main objective was met by dividing the research into sub-tasks, each with its own objective. The first objective was to develop, implement, and verify an MIMD molecular dynamics model which can be scaled to a variable number of processors. The second objective was to use this baseline implementation to solve a complex molecular dynamics problem of physical interest. The third objective was to analyze the baseline implementation and develop an analytical model for identifying the constraints to even faster execution. The results of the analysis, coupled with the practical experience gained from solving an actual problem, was used to develop a set of design criteria. The final system was designed to optimize the features from the design criteria.