Optimisation of data locality in energy calculations for large-scale molecular dynamics simulations

The main bottleneck of molecular dynamic simulations is the estimation of nonbonded pairwise interaction, which often employs neighbour search algorithms to find out interacting atom pairs. These methods have some drawbacks in fulfilling data locality principle, which is unable to take full advantage of modern computer architecture. In this article, we developed a new method by introducing a temporary list to reduce the sparsity in data access. This list permits to obtain a compact and sequential data structure which benefits to efficiently fulfil the data locality principle. We tested and compared the performance of the new method with that of the extensively used reordering method. The new method based on linked cell list is shown to increase 13% of computation speed and have better parallelism in comparison with reordering method. The increase in parallel efficiency makes the new method a promising option for large-scale molecular simulations.