31st JSST Annual Conference
International Conference on Simulation Technology
Sep. 27-28, 2012
Integrated Research Center, Kobe Univ., Japan
|Speaker||Prof. Mitsuo Yokokawa (Riken)|
|Title||The K computer and first results by trial access|
|Date||11:30-12:10 (Sep. 27)|
The K computer is a large-scale parallel supercomputer system with distributed memory. The system was started to develop as a seven-year national project in 2006, and was completed to develop in June, 2012. It will be operational and open to public at the end of September, 2012. The K computer is the first system which broke a 10 peta-FLOPS wall in LINPACK benchmark in November, 2011.
The main part of the K computer is composed of 82,944 compute nodes and 5,184 I/O nodes, and each node is mainly composed of a CPU, a set of memory with 16 gigabytes, and an LSI for interconnect between the nodes. The CPU is SPARC64 VIIIfx which has 8 cores and a 6-megabyte L2 cache shared by the cores on an LSI chip operating at a clock frequency of 2GHz and its peak performance is 128giga-flops. The nodes are connected by the newly developed interconnect network named Tofu which is a 6-dimensional mesh/torus network for data communication among compute nodes.
The operating system for the nodes is Linux. The programming languages of Fortran, C and C++ are available for users as a conventional users’ programming environment.
Part of CPU cycles has been provided to limited users as a trial since April, 2011. More than 30 applications were already optimized for the K computer and achieved more than 20000 parallelisms.
In this talk, outline of K computer and some results obtained by the early access will be given.
|Speaker||Prof. Takuya Matsuda (NPO Einstein, the Venture Laboratory, Kyoto University)|
|Coauthor||Hiromu Isaka (Shimadzu co.)|
|Title||Boltzmann Particle Hydrodynamics Methods and its Application to Cosmic Gas Flow|
|Date||14:45-15:25 (Sep. 27)|
In the present talk, we introduce various particle schemes for fluid dynamics, and then we discuss particularly the Boltzmann Particle Hydrodynamics method (BPH) proposed by the present author, which is suitable to simulate high speed gas flow in astrophysics. In the BPH scheme, we consider gas as a collection of simulated molecular particles as is in the DSMC. The aim of the method is to solve continuum flow rather than rarefied gas. The scheme is consisted by two steps: 1) we solve the Newtonian equation of motion of particles, 2) we assume that particles in a cell relaxes into a thermodynamic equilibrium state. By repeating the above two steps, we effectively solve the time evolution of either the Euler or the Navier-Stokes equation by solving the BGK equation, which is a simplified version of the Boltzmann equation. The merit of the BPH is its robustness: i.e. the time step is not restricted by the CFL condition although the scheme is time explicit. It can easily solve the motion of gas of very high (even infinitely high) Mach number. It can easily handle the presence of vacuum and time varying boundary conditions because of its particle nature. One demerit of the BPH is the fact that it requires large memory and CPU time, but the demerit will be shortly solved by a development of computers. We present some test problems and astrophysical flows. We also discuss a method of parallelization of the scheme.
|Speaker||Prof. Zhi-Wei Luo (Kobe Univ.)|
|Title||Innovation On Human Interactive Dynamic Simulation for Health Engineering|
|Date||11:30-12:10 (Sep. 28)|
The next generation of robotics and virtual reality are highly expected as core technologies in health engineering for the coming aging society to support and further promote human health in our everyday live. By now, the robotics was developed in such a way so that to separate and reduce the human activities from the automated production lines. However, the next generation of robot is also expected as human friendly partners to promote human’s activities and abilities in their home or street. In order to design and evaluate these human interactive technologies, development of real time dynamic simulation is important. Such dynamic simulation includes not only modeling and simulation of human motor function itself, but also complex physical and mental human-robot interactions. This talk will review the latest human motion analysis technologies, it then show the basic framework on dynamic simulation of human-robot interaction. As examples, it shows the study on robotic human care tasks and evaluation of high order brain functions of human. The talk also points out research subjects which are still left for us to challenge for human interaction dynamic simulation, such as multi-physical and multi-scale human simulation, reality of force interaction, and how such simulation related to the promotion of human’s health, et.al.