December 24(Wednesday), 2008
Organized by Global Scientific Information and Computing Center,
Tokyo Institute of Technology
Supported by JSPS Global COE program "Computationism as a Foundation for the Sciences".
JST CREST gULP-HPC: Ultra Low Power, High-Performance Computing via Modeling and Optimization of Next Generation HPC Technologiesh and
Venue Digital Multi-purpose Hall, W9-bld. [map]
Tokyo Institute of Technology [map]
Contact Takayuki Aoki (GSIC, Tokyo Tech)
e-mail: fifth-gsic-symposium@sim.gsic.titech.ac.jp

Scope

Exponential growth of the supercomputer makes it possible to open the next generation of the Computational Mechanics. New numerical methods, algorithms and devices should be taken into consideration there. In the symposium, the talks on leading studies of fluid and structural analyses, a new device computing and new algorithm for large matrix are arranged, and the perspective view of the future direction will be given through the symposium.

Program

13:00 Opening Address
Osamu WATANABE (Director of GSIC, Tokyo Institute of Technology)
13:10-14:10 [Keynote] Space-Time Fluid-Structure Interaction Modeling
Tayfun E. TEZDUYAR (Rice University)
14:10-14:50 Large Scale Structural Analysis \ The Present Situation and Future Prospects
Hiroshi AKIBA (Allied Engineering Corporation)
14:50-15:30 High-order schemes with multi-moment constraints
Feng XIAO (Tokyo Institute of Technology)
15:30-15:50 break
15:50-16:30 GPGPU Computing for CFD Applications as a SIMD-type Accelerator
Takayuki AOKI (Tokyo Institute of Technology)
16:30-17:10 Strategy of Domain Partitioning for Parallel Preconditioned Iterative Solvers in the Multi-Core Era
Kengo NAKAJIMA (University of Tokyo)

Abstract

Space-Time Fluid-Structure Interaction Modeling
Tayfun E. Tezduyar
Team for Advanced Flow Simulation and Modeling (T*AFSM)
Mechanical Engineering, Rice University

Abstract
The space-time fluid-structure interaction (FSI) techniques developed by the Team for Advanced Flow Simulation and Modeling (T*AFSM) have been applied to a wide range of 3D computation of FSI problems, some as early as in 1994 and many with challenging complexities. In this presentation, we review these space-time FSI techniques and describe the enhancements introduced recently by the T*AFSM to increase the scope, accuracy, robustness and efficiency of these techniques. The aspects of the FSI solution process enhanced include the Deforming-Spatial-Domain/Stabilized Space-Time (DSD/SST) formulation, the fluid-structure interface conditions, smoothing techniques in data exchange between the fluid and structure, and a contact algorithm protecting the quality of the fluid mechanics mesh between the structural surfaces coming into contact. We present a number of 3D numerical examples computed with these new space-time FSI techniques, including the FSI modeling of the new parachutes to be used for NASA's Orion space vehicle.

 

Large Scale Structural Analysis@\ The Present Situation and Future Prospects
Hiroshi Akiba
Allied Engineering Corporation

Abstract
In recent years, HPC (High Performance Computing) is rapidly growing. The fastest computer system in the world has achieved 1PFLOPS. The parallel computing environment has been changing from flat architecture to heterogeneous one. In computational mechanics, however, our effort to structural analysis, which we hope to benefit from HPC, is a little behind from the progress of HPC today. In this talk, we will look at how the structural analysis based on domain decomposition method works on the HPC environment. Also we will look at analysis examples and future direction of structural analysis on HPC.

 

High-order schemes with multi-moment constraints
Feng Xiao
Department of Energy Sciences,
Tokyo Institute of Technology

Abstract
A new type of high-order method based on local reconstruction is presented. The method uses the point values distributed within single mesh cell as the unknowns which are predicted in time. The time evolution equations used to update the unknowns are derived from a set of constraint conditions imposed on multi kinds of moments, i.e. the cell-averaged value and the point-wise value of the state variable and its derivatives. The finite volume constraint on the cell-average guarantees the numerical conservativeness of the method.  A multi-moment constrained (MC) Lagrange polynomial of any demanded order of accuracy can be constructed over the single cell and converts the evolution equations of the moments to those of the unknowns. A formulation with equally spaced points will be presented to show the substantial difference between a conventional Lagrange polynomial and an MC Lagrange polynomial in approximating the Runge function.

 

GPGPU Computing for CFD Applications as a SIMD-type Accelerator
Takayuki Aoki
Global Scientific Information and Computing Center
Tokyo Institute of Technology

Abstract
GPU (Graphics Processing Unit) has remarkably evoluted with high performance for GP (General Purpose) use. Especially, the NVIDIA CUDA (Compute unified device architecture) programming environment makes it easier to develop applications on GPUs. Some applications of CFD (Computational Fluid Dynamics) have been developed for the GPUs: the growth of the compressible Rayleigh-Taylor instability, the Tsunami simulation including run-up, the incompressible flow around a cylinder run on the GPU. These CFD applications are accelerated several 10 times faster than those executed on CPUs. It is found that the GPU computing is a promising direction for the next-generation CFD.

 

Strategy of Domain Partitioning for Parallel Preconditioned Iterative Solvers in the Multi-Core Era
Kengo Nakajima
Information Technology Center, University of Tokyo

Abstract:
Appropriate domain partitioning is the most critical and important issues for convergence of parallel preconditioned iterative solvers in large-scale real-world applications with ill-conditioned coefficient matrices on massively parallel computers. In this talk, HID (Hierarchical Interface Decomposition) [Henon & Saad 2007] and "Extended HID" [Nakaima 2008] will be described and preliminary results by 512 cores of T2K Open Supercomputer (Todai Combined Cluster) (T2K/Tokyo) are also shown. Comparison of Hybrid and flat MPI parallel programming models for HID on T2K/Tokyo is also discussed. Furthermore preliminary results of parallel multigrid method with HID will be introduced.