Achieving Scalable Parallel Molecular Dynamics Using Dynamic Spatial Domain Decomposition Techniques

To achieve scalable parallel performance in molecular dynamics simulations, we have modeled and implemented several dynamic spatial domain decomposition algorithms. The modeling is based upon the bulk synchronous parallel architecture model (BSP), which describes supersteps of computation, communication, and synchronization. Using this model, we have developed prototypes that explore the differing costs of several spatial decomposition algorithms and then use this data to drive implementation of our molecular dynamics simulator,Sigma. The parallel implementation is not bound to the limitations of the BSP model, allowing us to extend the spatial decomposition algorithm. For an initial decomposition, we use one of the successful decomposition strategies from the BSP study and then subsequently use performance data to adjust the decomposition, dynamically improving the load balance. The motivating reason to use historical performance data is that the computation to predict a better decomposition increases in cost with the quality of prediction, while the measurement of past work often has hardware support, requiring only a slight amount of work to modify the decomposition for future simulation steps. In this paper, we present our adaptive spatial decomposition algorithms, the results of modeling them with the BSP, the enhanced spatial decomposition algorithm, and its performance results on computers available locally and at the national supercomputer centers.