Laboratory of High-performance Computing
Towards the Computing Frontier
Our research focuses on high-performance computing (HPC), including the use of state-of-the-art supercomputers, cloud computing environments, many-core processors and GPUs. More concretely, we research parallel algorithms essential to the effective use of these high-performance computing environments and develop high-performance simulation programs, modelling and numerical computation libraries.
Our research on parallel algorithms and implementations focuses especially on matrix computations, including eigenvalue and linear system solvers. It’s well-known that Google’s PageRank is related to eigenvalues of matrices, and matrix computations are also frequently used in big data analysis and data mining. Matrix computations are also essential in fluid and structural analysis. Given the range of applications, there is strong demand for improvements in matrix computation. In our laboratory, we develop massively parallel multithreaded algorithms for many-core processors and matrix computation algorithms suitable for use on supercomputers such as the K computer. Some of these are included in commercial software packages and used outside academia.
In addition, we develop Jet FDTD, a high-performance, high-resolution system for analyzing massive electromagnetic fields using supercomputers. We use this system to perform large-scale computational simulations of indoor radiowave propagation. The picture gives a visualization of indoor propagation of a wireless LAN system. The particles show the extent to which the signal propagates, and also its strength.