Development of a Scalable Parallel Framework for Discrete Event Simulation

Taking advantage of the enormous computing power provided by ever-expanding clusters demands parallel algorithms that can scale to thousands of processes. Many systems, are, however, extremely difficult to solve in parallel because they involve an enormous number of events that occur in sequence. This project will develop and optimise a scalable parallel framework for performing discrete event simulation on a wide range of problems. While we will apply the framework to some aspects of image analysis and the modelling of fluid flow in porous materials, it can equally enable parallel simulation of a plethora of systems that arise in optimal control, cellular automata and other fields.

Principal Investigator

Adrian Sheppard
Applied Mathematics, RSPhysSE
Australian National University

Project

f99

Co-Investigators

Ji-Youn Arns
The School for Petroleum Eng.
University of NSW

Viet Nguyen
Mathematics, RSPhysSE
Australian National University

Gerd Schroeder
Boris Breidenbach
Applied Mathematics, RSPhysSE
Australian National University

RFCD Codes

280203, 280210

Significant Achievements, Anticipated Outcomes and Future Work

The time warp parallel framework that we developed previously received some, relatively minor, optimisations. The framework is currently used by three applications in the field of three dimensional image analysis: firstly, fast marching solution of the Eikonal equation, used for finding Euclidean distances and in segmentation and tessellation algorithms; secondly, distance-ordered homotopic thinning for skeletonisation and finally watershed transform for segmentation and general partitioning.

This year we have improved the efficiency of region growing algorithms in fast marching by identifying isolated regions that no longer need to be searched. This enabled our software to be applied very successfully for the segmentation of very high resolution "phase contrast" tomographic images obtained at the European Synchrotron Research Facility. This has allowed a quantitative, three dimensional, study of a range of biomaterials - including wood, bamboo and nut shells - at an unprecedented 200nm resolution. The accompanying image shows the isolated voids in the shell of a Macadamia nut.

The software development needs to continue, to enable the time-warp framework to run efficiently on cluster-type computers. there is an ever-increasing interest in this code from researchers around the world, many of whom only have access to clusters. A lack of human resources is preventing us from undertaking the development needed to reduce the number of messages that the time-warp framework exchanges. For this reason the project isn't being renewed in 2006.

 

Data Sources, Curation Techniques, Data Access Policy and Method

this project is primarily for software development and only has secondary access to any data from MDSS. See projects w09, d59 and x73 for more information.

 

Computational Techniques Used

This project looks to devise better communication algorithms for making use of a time-warp style parallel discrete event simulator in the context of three dimensional image processing.

 

Publications, Awards and External Funding

External Funding and Awards

None

Publications

None