Projects

The Centre for Parallel Computing has a number of research projects that investigate how parallel computing can be used to accelerate the scientific process. These projects aim to develop efficient implementations of scientific simulations and algorithms for the parallel devices at Massey University. The following are some brief descriptions of the simulations being investigated by the members of the CPC.

Name

Description

Members

Publications

N-Body

N-body simulations approximate the motion of a system of particles - often with some potential between them. These simulations can range in scale from interacting particles to stars and galaxies. Simulations may use a variety of numerical methods to approximate the particles' motion, they may also include collision detection/response models. The N-body project investigates the development of parallel implementations of N-body simulations. Efforts have been directed towards developing GPU and mGPU implementations of these simulations and investigating the associated challenges. Features investigated include:

  • Efficient GPU memory access patterns.
  • Spatial data structures.
  • GPU hard-sphere collision algorithms.
  • Communication schemes for mGPU simulations.

  • Daniel Playne
  • Ken Hawick
  • Alwyn Husselmann
  • Mitchell Johnson

CSTN-077
CSTN-120
CSTN-138
CSTN-139
CSTN-143
CSTN-156

Field Equations

Field equation models approximate the behaviour of a field as determined by a governing equation. The field is usually approximated by a discrete grid and the spatial terms of the equations by discrete stencils. Specific field equations investigated include: Cahn-Hilliard, Lotka-Volterra, Ginzburg-Landau and Heisenberg. This project explores methods of implementing field equation simulations on parallel devices including GPUs, mGPU systems, GPU-clusters, Cell Broadband Engines and multi-core machines.

  • Efficient GPU memory access for stencil operators.
  • Communication schemes for mGPU and GPU-cluster implementations.

  • Ken Hawick
  • Daniel Playne

CSTN-065
CSTN-070
CSTN-073
CSTN-074
CSTN-099
CSTN-111
CSTN-125
CSTN-146

Network Generation/Analysis

Complex Networks such as scale-free networks or small-world networks are necessary to model many systems in mathematics, physics, biology and computer science. Simulations using these networks can exhibit behaviour that is simply not seen on regular or random data structures. However, for large systems the generation and analysis of these networks presents a significant computational challenge. This project aims to investigate and develop efficient parallel algorithms and data storage patterns for use with these network structures. Features investigated include:

  • Parallel Network generation on GPUs and multi-core CPUs.
  • Parallel algorithms for network analysis.
  • Simulation of models on network structures.

  • Ken Hawick
  • Arno Leist
  • Chris Scogings

CSTN-117
CSTN-126
CSTN-149

Lattice-based Computational Models

Lattice-based computational models often show complex behaviour emerging from simple set of state and rules. The behaviour of these models can often mimic or approximate the behaviour of complex equations. Models investigated include: Lattice Gas, Ising, Potts, Sznajd and Game of Life. This project investigates how these lattice-based computational models can be efficiently implemented on parallel devices. Features investigated include:

  • Random Number Generation on parallel devices.
  • Bit-packing methods to reduce storage requirements and improve performance on memory-bound parallel devices.
  • Simulations of computational models on non-uniform data structures.

  • Ken Hawick
  • Daniel Playne
  • Arno Leist
  • Mitchell Johnson

CSTN-093
CSTN-104
CSTN-108
CSTN-109
CSTN-110
CSTN-135
CSTN-144
CSTN-148
CSTN-159

Scientific Visualisation

Visualisation is an important component of most scientific simulations. Correct visualisation of a simulation allows the behaviour of the simulation to be observed and are vital in identifying many important phenomena. Advanced visualisation methods often require a lot of preprocessing and data analysis to generate the visualisation. Parallel computing techniques can be used to perform this pre-processing in real time and allow the observer to interact with the simulation in real-time. Techniques Include:

  • Connected-Component Labelling.
  • Marching Cubes surface construction.

  • Daniel Playne
  • Ken Hawick

CSTN-110
CSTN-101
CSTN-082
CSTN-074