To achieve high-performance computing with K-computer, we need to use more than 80,000 computing nodes connected by a network so that they cooperate with each other by communicating data. However, the overall performance may be degraded by the big overhead for global communication and synchronization among nodes. In our research team, we develop computing accelerators to achieve large-scale computing with less performance degradation, by introducing a new parallel computing model based on localized communication and synchronization. This new approach contributes to advanced usage of K-computer and design space exploration of processors and networks for future supercomputers.
- Acceleration mechanisms for high-performance computing with a large-scale system
- Novel programming approach based on a data-flow computing model
- Development and evaluation of high-performance applications based on the acceleration mechanisms and the data-flow computing model