ACHIEVING SCALABLE PARALLEL MOLECULAR-DYNAMICS USING DYNAMIC SPATIAL DOMAIN DECOMPOSITION TECHNIQUES

To achieve scalable parallel performance in molecular dynamics simulat ions, we have modeled and implemented several dynamic spatial domain d ecomposition algorithms, The modeling is based upon the bulk synchrono us parallel architecture model (BSP), which describes supersteps of co mputation, communication, and synchronization. Using this model, we ha ve developed prototypes that explore the differing costs of several sp atial decomposition algorithms and then use this data to drive impleme ntation of our molecular dynamics simulator, Sigma. The parallel imple mentation is not bound to the limitations of the BSP model, allowing u s to extend the spatial decomposition algorithm, For an initial decomp osition, we use one of the successful decomposition strategies from th e BSP study and then subsequently use performance data to adjust the d ecomposition, dynamically improving the load balance, The motivating r eason to use historical performance data is that the computation to pr edict a better decomposition increases in cost with the quality of pre diction, while the measurement of past work often has hardware support , requiring only a slight amount of work to modify the decomposition f or future simulation steps, In this paper, we present our adaptive spa tial decomposition algorithms, the results of modeling them with the B SP, the enhanced spatial decomposition algorithm, and its performance results on computers available locally and at the national supercomput er centers. (C) 1997 Academic Press.