SC'98 High Performance Computing Challenge

Industrial Mold Filling Simulation
Using an Internationally Distributed
Software Component Architecture

This project connects together high performance software and hardware systems to provide an integrated solution environment for a 3D parallel finite element code modeling industrial material processes such as casting and injection molding. It encompasses problems of HPC resource allocation and management, component systems for problem-solving environments, parallel solution of implicit finite element systems on complex geometries, solution of large-scale distributed systems of equations, and 3D immersive visualization. One goal of the project is to demonstrate a complete, end-to-end high-performance solution of problems of major importance to industry, all occuring in an internationally distributed hardware and software environment.

Solve and visualize a large-scale industrial mold-filling process using internationally distributed software/hardware/human resources requires innovations in several areas:

Developing parallel solution techniques for time-dependent, multi-physics finite element simulations on complex geometries.
Creating user-level tools giving access to Globus infrastructure including GRAM and MDS) for organizing, finding, and instantiating software components as well as managing hardware resources.
Building the Component Architecture Toolkit (CAT), a compositional tool to orchestrate the software and hardware components for this and other distributed applications.
Developing suitable supporting HPC++ wrapped applications for the visualization and sparse linear solution components of the demonstration
Using Linear System Analyzer (LSA) software components to dynamically select solution strategies for the application.
Extending a practical 3D visualization tool (VU) to handle large-scale immersive visualization for scientific and engineering problems.

The Application

The driving application is a 3D parallel finite element code for modeling industrial material processes such as casting and injection molding. The CASTVIEW code has been developed at the Industrial Materials Institute (IMI/NRC) in Quebec. CASTVIEW uses Metis for partitioning the algebraic systems and MPI for parallel assembly and solution of large sparse linear systems. This application is particularly challenging for HPC parallel methods, because it

models multiple quantities which have extreme variations in magnitudes
models tightly coupled physics which require fully implicit methods
handles engineering problems with complex geometries
has a dynamically changing workload
solves a wide variety of problems in research engineering, and implements a large variety of mathematical models and numerical methods

For SC98 two problems are demonstrated. The first is a 2.5M element Pratt & Whitney engine part, and the second is a smaller engine piston. A sequence of images showing the piston simulation is available.

SC98 Demonstrations

Note that some of the demonstrations are simultaneous.

Location	Date	Time
iGrid Booth R130	Monday, 9 November	19:00 - 21:00
NCSA Alliance Booth	Tuesday 10 November	13:30 - 14:45
SGI CAVE Booth	Tuesday 10 November	13:00 - 14:00
iGrid Booth R130	Wednesday, 11 November	16:30 - 17:00
SGI Cave Booth	Wednesday, 11 November	15:00 - 16:00
NCSA Booth and SGI CAVE	Thursday, 12 November	12:00 - 13:00

Participants

Relevant Technologies

Framework:
Finite element code: CASTVIEW, developed at Industrial Materials Institute (IMI), National Research Council of Canada, Boucherville, Québec, Canada.
Parallel solver from Sandia National Lab by Shadid and Tuminaro: Aztec
Visualization: VU, developed at the Center for Research on Computation and its Applications (CERCA), Montréal, Québec, Canada.
Software Resource Information Services: MDS Software Schema, which in turn is based on the use of LDAP protocols.
CAT Graphical User Interface: CAT UI

bramley@cs.indiana.edu