Browsing by Author "Hawick, K.A."

Now showing 1 - 13 of 13
  • Thumbnail Image
    Item
    64-bit architechtures and compute clusters for high performance simulations
    (Massey University, 2006) Hawick, K.A.; James, H.A.; Scogings, C.J.
    Simulation of large complex systems remains one of the most demanding of high performance computer systems both in terms of raw compute performance and efficient memory management. Recent availability of 64-bit architectures has opened up the possibilities of commodity computers accessing more than the 4 Gigabyte memory limit previously enforced by 32-bit addressing. We report on some performance measurements we have made on two 64-bit architectures and their consequences for some high performance simulations. We discuss performance of our codes for simulations of artificial life models; computational physics models of point particles on lattices; and with interacting clusters of particles. We have summarised pertinent features of these codes into benchmark kernels which we discuss in the context of wellknown benchmark kernels of the 32-bit era. We report on how these these findings were useful in the context of designing 64-bit compute clusters for high-performance simulations.
  • Thumbnail Image
    Item
    Accelerated face detector training using the PSL framework
    (Massey University, 2009) Susnjak, T.; Barczak, A.L.C.; Hawick, K.A.
    We train a face detection system using the PSL framework [1] which combines the AdaBoost learning algorithm and Haar-like features. We demonstrate the ability of this framework to overcome some of the challenges inherent in training classifiers that are structured in cascades of boosted ensembles (CoBE). The PSL classifiers are compared to the Viola-Jones type cas- caded classifiers. We establish the ability of the PSL framework to produce classifiers in a complex domain in significantly reduced time frame. They also comprise of fewer boosted en- sembles albeit at a price of increased false detection rates on our test dataset. We also report on results from a more diverse number of experiments carried out on the PSL framework in order to shed more insight into the effects of variations in its adjustable training parameters.
  • Thumbnail Image
    Item
    A framework and simulation engine for studying artificial life
    (Massey University, 2004) James, H.A.; Scogings, C.; Hawick, K.A.
    The area of computer-generated artificial life-forms is a relatively recent field of inter-disciplinary study that involves mathematical modelling, physical intuition and ideas from chemistry and biology and computational science. Although the attribution of “life” to non biological systems is still controversial, several groups agree that certain emergent properties can be ascribed to computer simulated systems that can be constructed to “live” in a simulated environment. In this paper we discuss some of the issues and infrastructure necessary to construct a simulation laboratory for the study of computer generated artificial life-forms. We review possible technologies and present some preliminary studies based around simple models.
  • Thumbnail Image
    Item
    Managing community membership information in a small-world grid
    (Massey University, 2005) Hawick, K.A.; James, H.A.
    As the Grid matures the problem of resource discovery across communities, where resources now include computational services, is becoming more critical. The number of resources available on a world-wide grid is set to grow exponentially in much the same way as the number of static web pages on the WWW. We observe that the world-wide resource discovery problem can be modelled as a slowly evolving very-large sparse-matrix where individual matrix elements represent nodes’ knowledge of one another. Blocks in the matrix arise where nodes offer more than one service. Blocking effects also arise in the identification of sub-communities in the Grid. The linear algebra community has long been aware of suitable representations of large, sparse matrices. However, matrices the size of the world-wide grid potentially number in the billions, making dense solutions completely intractable. Distributed nodes will not necessarily have the storage capacity to store the addresses of any significant percentage of the available resources. We discuss ways of modelling this problem in the regime of a slowly changing service base including phenomena such as percolating networks and small-world network effects.
  • Thumbnail Image
    Item
    Mixing multi-core CPUs and GPUs for scientific simulation software
    (Massey University, 2010) Hawick, K.A.; Leist, A.; Playne, D.P.
    Recent technological and economic developments have led to widespread availability of multi-core CPUs and specialist accelerator processors such as graphical processing units (GPUs). The accelerated computational performance possible from these devices can be very high for some applications paradigms. Software languages and systems such as NVIDIA's CUDA and Khronos consortium's open compute language (OpenCL) support a number of individual parallel application programming paradigms. To scale up the performance of some complex systems simulations, a hybrid of multi-core CPUs for coarse-grained parallelism and very many core GPUs for data parallelism is necessary. We describe our use of hybrid applica- tions using threading approaches and multi-core CPUs to control independent GPU devices. We present speed-up data and discuss multi-threading software issues for the applications level programmer and o er some suggested areas for language development and integration between coarse-grained and ne-grained multi-thread systems. We discuss results from three common simulation algorithmic areas including: partial di erential equations; graph cluster metric calculations and random number generation. We report on programming experiences and selected performance for these algorithms on: single and multiple GPUs; multi-core CPUs; a CellBE; and using OpenCL. We discuss programmer usability issues and the outlook and trends in multi-core programming for scienti c applications developers.
  • Thumbnail Image
    Item
    Node importance ranking and scaling properties of some complex road networks
    (Massey University, 2007) Hawick, K.A.; James, H.A.
    The scaling and other quantifiable properties of a network have recently been proven valuable in understanding the robustness and vulnerability properties of various societal and infrastructural networks. In this paper we revisit the algorithms for computing various quantifiable properties of a planar road network and consider the algorithmic complexity and scalability in the light of recent technological advances. We compute properties for a sample of interesting trunk road networks and discuss their applicability in determining the relative importance or criticality to the whole network of a particular node. We discuss the implications of present and anticipated technological capabilities in calculating properties for anticipated network sizes in the light of 64-bit computer architectures and commodity parallel computing.
  • Thumbnail Image
    Item
    A novel bootstrapping method for positive datasets in cascades of boosted ensembles
    (Massey University, 2010) Susnjak, T.; Barczak, A.L.C.; Hawick, K.A.
    We present a novel method for efficiently training a face detector using large positive datasets in a cascade of boosted ensembles. We extend the successful Viola-Jones [1] framework which achieved low false acceptance rates through bootstrapping negative samples with the capability to also bootstrap large positive datasets thereby capturing more in-class variation of the target object. We achieve this form of bootstrapping by way of an additional embedded cascade within each layer and term the new structure as the Bootstrapped Dual-Cascaded (BDC) framework. We demonstrate its ability to easily and efficiently train a classifier on large and complex face datasets which exhibit acute in-class variation.
  • Thumbnail Image
    Item
    Parallel containers: a tool for applying parallel computing applications on clusters
    (Massey University, 2004) Gan-El, M.; Hawick, K.A.
    Parallel and cluster computing remain somewhat difficult to apply quickly for many applications domains. Recent developments in computer libraries such as the Standard Template Library of the C++ language and the Message Passing Package associated with the Python Language provide a way to implement very high level parallel containers in support of application programming. A parallel container is an implementation of a data structure such as a list, or vector, or set, that has associated with it the necessary methods and state knowledge to distribute the contents of the structure across the memory of a parallel computer or a computer cluster. A key idea is that of the parallel iterator which allows a single high level statement written by the applications programmer to invoke a parallel operation across the entire data structure’s contents while avoiding the need for knowledge of how the distribution is actually carried out. This transparency approach means that optimised parallel algorithms can be separated from the applications domain code, maximising reuse of the parallel computing infrastructure and libraries. This paper describes our initial experiments with C++ parallel containers.
  • Thumbnail Image
    Item
    A review of traffic simulation software
    (Massey University, 2009) Kotusevski, G.; Hawick, K.A.
    Computer simulation of tra c is a widely used method in research of tra c modelling, planning and development of tra c networks and systems. Vehicular tra c systems are of growing concern and interest globally and modelling arbitrarily complex tra c systems is a hard problem. In this article we review some of the tra c simulation software applications, their features and characteristics as well as the issues these applications face. Additionally, we introduce some algorithmic ideas, underpinning data structural approaches and quanti able metrics that can be applied to simulated model systems.
  • Thumbnail Image
    Item
    Simulating large random Boolean networks
    (Massey University, 2007) Hawick, K.A.; James, H.A.; Scogings, C.J.
    The Kauffman N-K, or random boolean network, model is an important tool for exploring the properties of large scale complex systems. There are computational challenges in simulating large networks with high connectivities. We describe some high-performance data structures and algorithms for implementing large-scale simulations of the random boolean network model using various storage types provided by the D programming language. We discuss the memory complexity of an optimised simulation code and present some measured properties of large networks.
  • Thumbnail Image
    Item
    Simulation modelling and visualisation: toolkits for building artificial worlds
    (Massey University, 2008) Playne, D.P.; Gerdelan, A.P.; Leist, A.; Scogings, C.J.; Hawick, K.A.
    Simulations users at all levels make heavy use of compute resources to drive computational simulations for greatly varying applications areas of research using different simulation paradigms. Simulations are implemented in many software forms, ranging from highly standardised and general models that run in proprietary software packages to ad hoc hand-crafted simulations codes for very specific applications. Visualisation of the workings or results of a simulation is another highly valuable capability for simulation developers and practitioners. There are many different software libraries and methods available for creating a visualisation layer for simulations, and it is often a difficult and time-consuming process to assemble a toolkit of these libraries and other resources that best suits a particular simulation model. We present here a break-down of the main simulation paradigms, and discuss differing toolkits and approaches that different researchers have taken to tackle coupled simulation and visualisation in each paradigm.
  • Thumbnail Image
    Item
    Small-world networks, distributed hash tables and the e-resource discovery problem
    (Massey University, 2010) Leist, A.; Hawick, K.A.
    Resource discovery is one of the most important underpinning problems behind producing a scalable, robust and efficient global infrastructure for e-Science. A number of approaches to the resource discovery and management problem have been made in various computational grid environments and prototypes over the last decade. Computational resources and services in modern grid and cloud environments can be modelled as an overlay network superposed on the physical network structure of the Internet and World Wide Web. We discuss some of the main approaches to resource discovery in the context of the general properties of such an overlay network. We present some performance data and predicted properties based on algorithmic approaches such as distributed hash table resource discovery and management. We describe a prototype system and use its model to explore some of the known key graph aspects of the global resource overlay network - including small-world and scale-free properties.
  • Thumbnail Image
    Item
    Sparse cross-products of metadata in scientific simulation management
    (Massey University, 2005) James, H.A.; Hawick, K.A.
    Managing scientific data is by no means a trivial task even in a single site environment with a small number of researchers involved. We discuss some issues concerned with posing well-specified experiments in terms of parameters or instrument settings and the metadata framework that arises from doing so. We are particularly interested in parallel computer simulation experiments, where very large quantities of warehouse-able data are involved. We consider SQL databases and other framework technologies for manipulating experimental data. Our framework manages the the outputs from parallel runs that arise from large cross-products of parameter combinations. Considerable useful experiment planning and analysis can be done with the sparse metadata without fully expanding the parameter cross-products. Extra value can be obtained from simulation output that can subsequently be data-mined. We have particular interests in running large scale Monte-Carlo physics model simulations. Finding ourselves overwhelmed by the problems of managing data and compute ¿resources, we have built a prototype tool using Java and MySQL that addresses these issues. We use this example to discuss type-space management and other fundamental ideas for implementing a laboratory information management system.