University of Newcastle upon Tyne | Faculty of SAgE | School of Computing Science
Parallelism Research Group University of Newcastle upon Tyne

Objectives

Sorry but these pages are not currently being maintained

Harnessing parallelism in the interests of greater - and scaleable - performance remains one of the key challenges of computing research. The main focus of research in the Parallelism group is addressing this challenge, particularly though a focus on object-oriented parallel systems.

Many application areas are turning to objects as a way of structuring their future data/information capture needs (e.g. bioinformatics, geoinformatics, physics). The current focus of our research is predicated on:

  1. The growing importance of object-orientation
  2. The need for object-oriented databases
  3. The need to exploit parallelism to achieve desired performance levels for large applications.

The Grid is becoming increasingly important as an infrastructure to support the E-scientist, providing the (parallel) computing and data resources necessary to enable the e-scientist to use, generate and share information. Helping to provide and support this infrastructure, and conduct research into the way it can be used, are also important parts of the research undertaken in the Newcastle group.

The research interests of the parallelism group span a number of interrelated areas. The diagram below captures the main areas of interest, and links to the following main areas of interest:

Research Interests diagram

Applications Tools Runtime support Platforms Objectives

Parallel platforms

A variety of parallel processing platforms exist. Architecturally, there are shared-memory multiprocessor systems and distributed-memory multiprocessor systems, each with different performance characteristics. Parallel platforms can be constructed from "commodity" hardware, such as PC clusters linked by high speed networks, or may be more specialist high-performance systems at the top-end of the performance (and price) range.

Details of the present (and past) platforms used by the Newcastle group can be found here. Our research interests are not in developing different architectures and platforms, but in providing higher-level software layers that, ideally, can hide the characteristics of the different platforms from the end-user.

Run-time systems

Run-time support can hide architectural differences and provide a uniform model to higher-level tools. Two particular areas of interest of the Newcastle group are run-time support for parallel, object-oriented systems, and Grid middleware support.

PhD-led work on run-time systems has resulted in the design and implementation of a novel system (NIP) that advances the state-of-the-art in two main areas: lazy task creation, and object-based distributed shared memory.

The North East Regional e-Science Centre has recently been established at Newcastle, funded by the EPSRC and DTI, to promote the use of the rapidly expanding capabilities of the Grid. Our research interests are in collaborating with the national and international efforts at implementing the Grid vision in general, and in particular in the areas of research concerning the provision of grid-enabled databases.

Tools

Designing and implementing parallel software systems remains an intellectually challenging task, being addressed at Newcastle by research into visual programming languages. PhD research has investigated visual dataflow languages, most recently in the development of Vorlon which was an object-oriented visual language. Leading on from the Vorlon and NIP research, the EPSRC-funded HiPPO (High Performance Parallel Object-oriented) project has the design and implementation of a visual language for parallel, object-oriented software as its main goal.

Parallel, object-oriented databases are likely to be a valuable tool for the e-scientist, and the design and implementation of such databases and their availability in Grid-based environments are key research interests of the group. The world's first parallel, object-oriented database server was developed jointly by the Universities of Manchester and Newcastle in the Polar project. A follow-on project named Polar 2 has also been funded by the ESPRC to investigate the problem of grid-based database query optimisation. The MyGrid project is also concerned with grid-based databases.

Applications

Parallelism is one of many areas where real experience from executing software often provides results worse than expected! The group is keen to work in application areas that can act as a test-bed for the technology developed as well as enhance experience. Applications encompassing numerical algorithms, virtual reality, parallel modelling and simulation are all areas of interest. More recent interest is in the area of Bioinformatics, an area with rapidly growing needs for computing and data resources. The HiPPO language will be used to develop applications covering a number of these areas. The MyGrid project has bioinformatics as its main application area.

Current work in the numerical area is concerned with parallel, object-oriented numerical libraries, through PhD research and through the HiPPO project.

An important application research area is in mathematical modelling, simulation and optimization of parallel and distributed systems. A novel parallel simulation algorithm (originally developed and patented by Prof. Mitrani and colleagues at AT&T Bell Laboratories) was generalized and applied successfully to the simulation of ATM switches and TCP flow control. Several collaborations with INRIA, the University of Denver, and the Department of Mathematics and Statistics at Newcastle have led to a number of advances in optimal threshold policies and heavy traffic asymptotics. Projects have been funded by British Telecom to study performance issues in Wide Area Networks. That work has already resulted in an optimal design for a distributed directory, and in the computation of low-cost policies for mobile agents. Current efforts are directed at the evaluation of performance trade-offs in reliable multicast protocols.