NOTUR Technology Transfer Projects TTP

1
NOTURTechnology Transfer Projects(TTP)

• By
• Trond Kvamsdal
• NOTUR
• 10. Juni 2004, Tromsø, Norway

2
CONTENTS

• The concept behind the TTPs
• Results obtained from the TTPs
• Concluding remarks

3
Purpose

• Enable optimal utilisation of the national high
  performance computing infrastructure

4
Bridging the adaptability gap
Peak performance
Performance
Application performance
Time
5
Main keywords

• User driven
• Strongly focused
• Transfer of exploitable technology
• Multi-disciplinary technology development
• National dimension
• Vertical integration

6
Computational Science and Engineering
MAC-model
Application
CSE
Mathematics
ComputerScience
7
The projects Themes

• TTP1 Scalable Algorithms
• TTP2 Scientific Visualisation
• TTP3 Climate and Environment
• TTP4 Medicine
• TTP5 Nature Science

8
TTP1 SALSAScalable algorithms for solving large
scale sparse linear systems
Tubular joint
Domain Decomposition
Electrical activity of the heart

• Partners UiB, NTNU, Simula, DNV

9
TTP1 Results
end-user application
- SALSA 1.0 made available to project partners
(01/2003) for AIX, IRIX - SALSA 2.0 (alpha)
available on request (general www release
2H2004) - final software requires further testing
before general release


SALSA
MUMPS
- software generally available (www) more than
100 downloads (version 4.3.2) - international
collaboration Parallab, CERFACS, RAL, INRIA,
ENSEEIHT-IRIT - used as local solver in domain
decomposition schemes from TTP3 - embedded in
SALSA, interfaces in latest PETSc (2.2.0,
02/2004) - tested/used on AIX, IRIX, Linux,
Solaris

MC64
- MC64 is (licensed) code in Harwell Subroutine
Library (Koster Duff, 1999-2004) - new MC64
available on request (general www release
1H2004) - embedded in (parallel) direct solvers
MUMPS and SuperLU (Li, Demmel, Gilbert) - proven
to accelerate various (preconditioned) iterative
solution schemes - tested/used on AIX, IRIX,
Linux, HP-UX, Solaris
MC64 gtgt MUMPS gtgt SALSA
10
TTP2 DINVISDistributed Network Visualisation

• Partners Ceetron, NTNU, UiO, UiT, DNMI, SINTEF

11
The Visualization Application

• High speed visualization of Scientific Data
• Easy integration with in-house codes
• API available for customization of visualization
  functionality and GUI
• Distributed, collaborative visualization obtained
  using SGI VizServer
• Available to all academic users of the NOTUR
  consortium

12
Data Compression
13
TTP3 Parallel local wind model
HIRLAM h10 km
MC2 h1 km
SIMRA h0.1 km
Værnes Airport

• Partners NTNU, UiB, DNMI, SINTEF

14
Example on results
Turbulence along landing path for aeroplanes at
Trondheim Airport (Værnes)
Achieved speedup in local code
15
TTP4 NEUROINFHigh Performance Computing for
Structural Neuroinformatics

• Partners UiO, NTNU, UiB, SINTEF

16

• The Java3D based client can run as web applet or
  stand-alone application, and communicates with a
  server over tcp/ip using a specially designed
  protocol.
• The server is separated into two parts a front
  end for client requests and a back-end for
  parallel computations.
• The server can run on
• a single-cpu
• smp machines, and
• mpi clusters.
• 1) and 2) are currently operative, preparations
  made for 3)
• Important new capability tools for generating
  and editing 3-D geometry from scalar fields,
  relevant for surface rendering from large data
  sets representing brain regions and neural
  networks.

17
TTP5 DUMADevelopment of user oriented modules
for the application of parallel algorithms in
nano and microscale simulation
Alanin
Two Droplets

• Partners UiB, NTNU, UiT

18
Project 1 Classical Molecular Dynamics

• - Development of framework
• - Integrators
• - Force Calculators
• - Parallelization

19
Project 2 Time Dependent Schrödinger Physics
20
Budget

• Activity Budget No of persons
• kNOK involved
• TTP1 5000 10
• TTP2 5000 13
• TTP3 4000 7
• TTP4 3500 13
• TTP5 3900 20

21
Concluding remarks

• The TTPs has met the expectation regarding
  efficient development of usable technology
• Better utilisation of the national
  HPC-infrastructure
• National impact obtained through collaboration