Patrick Sunter

High-performance computing provides unprecedented capabilities to produce higher resolution 4-D models in a fraction of time. Thus, the need exists for a new generation of visualization systems able to main- tain parity with the enormous volume of data generated. In attempting to write this much data to disk, each computational step introduces a significant performance bottleneck, yet most existing visualization software packages inherently rely on reading data in from a dump file. Available packages make this assumption of post- processing at quite a fundamental level and are not very well suited for plotting very large numbers of specialized particles. This necessitates the creation of a new visuali- zation system that meets the needs of large-scale geodynamic modeling. We have developed such a system, gLucifer, using a software framework approach that allows efficient reuse of our efforts in other areas of research. gLucifer is capable of producing movies of a 4-D data set ‘‘on the fly’’ (simultaneously with running the parallel scientific application) without creating a performance bottleneck. By eliminating most of the human efforts involved in visualizing results through postprocessing, gLucifer reconnects the scientist to the numerical experiment as it unfolds. Data sets that were previously very difficult to even manage may be efficiently explored and interrogated without writing to disk, and because this approach is based entirely on memory distributed across as many processors as are being utilized by the scientific application, the visualization solution is scalable into terabytes of data being rendered in real time.

Why is it that though the free software movement largely sprung from academic computer labs in the 1980s, some fields of scientific research have been sluggish in developing quality open source community codes? We'll examine the causes of this, and present an overview of the approaches we've employed to deal with them during the development of StGermain-to-Underworld, an open source scientific modelling framework.  While some of the issues discussed are specific to scientific computing, we hope many of them are broadly applicable to projects requiring a high degree of domain-specific knowledge and/or complex hardware and algorithms - such as games & mobile computing.

With the requirements of scientific software being continually extended, the need for fast, scalable numerical solvers is always of concern. Geometric multigrid has been shown to provide theoretically linear scalability for a wide range of mathematical problems, gaining popularity as an effective numerical solution technique. However, geometric multigrid is directly influenced by the size and parallel decomposition of a problem's spatial discretisation. The effect of one-dimensional and three-dimensional parallel spatial decompositions on a parallel multigrid solver is investigated in the context of the StGermain scientific software framework. Quantitative results are provided as a result of running a test problem on the APAC national facility.

There are many attempts at libraries and frameworks targeted at reducing the difficulty and cost of developing computational codes. To satisfy this goal they typically aim to reduce the amount of code needed to be written by a single scientist. Is it possible that we have out- grown this approach? The growing trend in successful research codes is the development of group and/or community efforts. In this case, individuals spanning departmental, financial and political borders contribute to the one greater effort and consequently leverage on a greater skills and experience base. Human labour is an expensive resource required for any software development, and so, for any consortium undertaking computational code development, it becomes important to address the way in which the community participates in such an activity. StGermain is a fundamental framework that attempts to eliminate specifically this problem. In this paper we discuss the issues of academic community coding efforts from a code maintainability point of view, and how StGermain provides programming constructs that address these issues.

There are two ways to interpret a title such as " A Plug-in based design for code maintainability in HPC " , based on whether you are: through and through a HPC traditionalist, and then there is everybody else. But with realisation of computer software development cost, the commoditisation of clustering and the requirement of cross disciplinary science, scientific code evolution and maintenance for researchers is a real issue. This paper investigates what performance costs does one bare for the flexibility and maintainability of HPC software. If we can't be fully flexible, what can be? With all modern HPC platforms facilitating expected features such as dynamic libraries, concepts such as plugins can be considered. We explain how and where and why such concepts may be utilised. Then we offer two indicative examples of two very differently formulated scientific codes.

Each discipline of geophysics has traditionally focused on limited sets of closely related phenomena using methodologies and data sets optimized for its specific area of interest. Why is that? Single discipline, single scale, foundation physics problems are relatively easy to code in Fortran, and hence they eventually become optimized for best performance whilst simultaneously becoming difficult to adapt to new

One of the important directions in computational E-Research is the drive towards “Reproducible Research” [1]. Briefly, this involves a move to publishing all the data, tools and steps needed to support scientific results communicated in the literature, so they can be reproduced by other researchers. The key claimed benefits are increasing the rigour of research by allowing computational & domain scientists to test the claims of others in their field, and their productivity by avoiding the need to re-construct the steps and data needed to perform a particular analysis. While open-source software and data repositories go part-way towards providing this capability, as argued in [1] a means of recording and communicating the series of computational and analysis steps that produced a result are also needed. In recent years several software systems have developed to support such a goal of recording and re-running reproducible scientific workflows, such as Kepler and Taverna, both open-source systems [2].

In computational modelling of physical phenomena in fields such as climate science and geophysics, one of the key areas where this concept of reproducible research can be usefully applied is in the well-established tradition of development, assessment and communication of ‘benchmarks in the literature’. Here 'benchmarking' can mean evaluating any facet of scientific features, numerical accuracy, or computational performance in terms of time, memory or convergence. Benchmarking as an activity generally requires both considerable computation resources, and a multi-stage workflow involving assembling and configuring models, running them, then analyzing and post-processing their results in various ways. In it’s latter stages it is amenable to Grid computing, but requires considerable interactive development of the analysis that is generally performed on local resources.

Starting from these two concepts, this talk will report and reflect on an effort to apply them in practice for a computational community developing and using the Underworld parallel HPC Geophysics application ([3,4]). This effort has been embodied by a new Python toolkit called CREDO that enables the creation of benchmarking scripts, and enhancements to the Bitten continuous integration toolkit that hosts the group’s codebases and testing results. Underworld is co-developed by Monash University, VPAC and several external contributors, with primary funding currently provided by the AuScope NCRIS program. Underworld utilizes the StGermain computational framework [5,6], and the current effort extends and builds upon pervious iterations in a suite of unit testing, analysis and visualisation tools developed by the group [7,8].

The radial return mapping algorithm within the computational context of a hybrid Finite Element and Particle-In-Cell (FE/PIC) method is constructed to allow a fluid flow FE/PIC code to be applied solid mechanic problems with large displacements and large deformations. The FE/PIC ...

In Australia the energy supplied by the stationary energy sector is mainly generated from non renewable sources such as coal, oil, petroleum fuels, natural gas or LPG (AGO 2006). In order to reduce greenhouse gas (GHG) emissions there needs to be a significant focus on adopting alternative renewable sources of energy. Targets for renewable energy are being set internationally, nationally and at state level which is placing pressure on local governments and their communities to increase the use of energy from renewable sources.

This paper identifies the renewable energy resources that exist within the City of Onkaparinga Local Government Area (LGA) in greater Adelaide, South Australia. It reviews the methods that have been used to assess renewable energy at local level and uses the approach used for the renewable energy assessment for the City of Playford (in South Australia) which was first developed in the United Kingdom to identify the Resource Base, the Resource, and the Reserve for each of solar energy, wind energy and biomass energy in the case study area. GIS methods were used to identify solar and wind resources across the case study area. The results of the assessment indicated that there are significant solar, wind and biomass resources in the Onkaparinga LGA, and that current economic conditions would allow a significant solar resource to be substituted for the energy currently being provided from non-renewable sources.