The large-scale parallelization of the Boids model on the K computer and the heterogeneous multi-processing unit

While attempting the parallelization of multiagent systems, a load imbalance often slows down the calculation. Hirokawa et al. presented a multilevel hierarchy of parallelism to mitigate the load-imbalance problem of the Boids model using a pseudo-quadruple arithmetic technique. In this paper, we modify the hierarchies of parallelization for both the K computer and a heterogeneous multiprocessing (HMP) unit, and evaluate the performance of the proposed solutions on these systems. We observed that the parallelization can decrease the negative effects of a load imbalance. Furthermore, we discovered that the power consumption can be effectively controlled by calculating the load and that the simultaneous use of both high-performance cores and low-energy cores slows down the calculations on the HMP unit. Parallelization has been reported to have the potential for enabling the completion of simulations in real time and to have the ability to evaluate up to 3000 billion agents using the K computer.