Beowulf

From SystemsResearch

Jump to: navigation, search

Contents

JREI: Distributed Software Systems

Overview

The EPSRC grant GR/M32351/01 was awarded by the Joint Research Equipment Initiative, for a distributed multi-computer to support the work of seven research projects within the School of Computer Science at the University of St Andrews. As will be described below, the distributed multi-computer (a 64 node Beowulf system) provided further support for a range of well established research already funded by EPSRC: GR/L32699 "Compliant Systems Architecture"; GR/L93379 "Granularity Analysis for Parallel Functional Programs"; GR/L21013 "The Development and Application of the GAP System"; and GR/L48256 "Computer Algebra and Automated Reasoning". Furthermore the equipment was used to support successful bids for new EPSRC funded research: GR/M74931 "Collecting Distributed Garbage using the DMOS Family of Algorithms"; GR/M88938 "Compliant Systems Architecture - Part 2" and EPSRC CASE award 99802449. Munro & Morrison have also been awarded Australian ARC Large Grant A10033014 for work on garbage collection that will use the Beowulf. Finally by providing a focus for our work the Beowulf allowed us to enhance our Visiting Fellow and Industrial programmes as well as providing a base for advanced postgraduate (PhD) training.

Our industrial partners in the research, Data Connection Ltd., made a cash grant of £10,000 towards the cost of the Beowulf computer in support of one aspect of the research. The University of St Andrews matched this sum from its endowment (non-science) funds and will also provide infrastructure and maintenance funding of £15,000 over the next three years. The configuration is shown below.

Image:Beowulf.gif

Objectives

The specific objectives of the proposal derive from the aims of the individual research projects. They were:

  • To develop techniques for predicting and measuring the scalability of existing commercial software systems which evolve from uni-processor to multi-computer implementations.
  • To target the EPSRC funded Compliant Systems Architecture (CSA) to a distributed multi-computer base. This will allow true concurrency and distribution and enable realistic measurement of the architectural paradigm.
  • To produce the first implementation of the distributed mature object space (DMOS) distributed garbage collector algorithm and to measure the effect of different policy mechanisms on its efficiency and scalability.
  • To develop algorithms and tools for practical algebraic computation on economical multi-computers.
  • To extend parallel Haskell with locality and clustering control mechanisms in order to support large irregular parallel applications on a loosely-coupled distributed-memory machine; and provide an instrumented implementation of these mechanisms in order to allow the rapid development of such algorithms.
  • To exploit cost-effective commodity based clusters for scalable service provision.
  • To explore parallelism in automated reasoning systems.

Summary

The Beowulf multi-computer will allow us to extend our research objectives by investigating the scalability of our systems and design on a class of parallel machine that is likely to dominate near-future, high-performance computing outside specialist computer centres. In addition, focusing our work on multi-computer machines will allow us to enhance our Visiting Fellow and Industrial programme as well as providing a base for advanced postgraduate (PhD) training. Our industrial partners in the bid are Data Connection Ltd who will make a grant of £10000 towards the cost of the distributed multi-computer in support of one aspect of the research. The University of St Andrews will match this sum from its endowment (non-science) funds and will also provide infrastructure and maintenance funding of £15000 over the next three years. Cost effective utilisation of EPSRC, University and Industrial resources will be achieved by sharing the proposed equipment among seven research projects. They are: Prediction and measurement of Scalability in Software Products; System Architectures for Process Modelling; Distributed Garbage Collection; Computational Algebra; Parallel Functional Languages; A Replicated resource Architecture for Scalable service provision and Safety and Correctness of Scientific Systems.

The major results of the research are:

  • about 30 published refereed papers related to the equipment; 7 technical reports; 2 books; collaboration with 20 visitors; 2 keynote addresses, and input into the EPSRC projects GR/M74931 "Collecting Distributed Garbage using the DMOS Family of Algorithms", GR/M88938 "Compliant Systems Architecture Part 2"; ARC Large Grant A10033014; and EPSRC CASE award 99802449.

Project Members

Funding

Funding Body EPSRC
Grant Number GR/M32351
Holders Ron Morrison, Kevin Hammond, Steve Linton, Graham Kirby, Colin Allison, Ursula Martin,Dave Munro
Value £74,826
Duration June 1999 - May 2000
Personal tools