K.A.Hawick
Research Data Network CRC & CRC for Advanced Computational Systems
Department of Computer Science, The University of Adelaide,
Australia SA 5005
Email khawick@cs.adelaide.edu.au
Tel +61 8 8303 4519, Fax +61 8 8303 4366
31 March 1998
Distributed and high-performance computing is an important area for computing and computational scientists in Australia. It is critical that the scarce high-performance resources available in Australia be made as widely accessible as possible to Australian scientists throughout the country. This is highly necessary for both present and likely future resources. Dedicated bandwidth is critical to some operations where it would be impossible to function in competition with the general public making World Wide Web (WWW) transactions across the same lines.
This project concerns demonstrating the utility of dedicated networked access to high-performance resources and quantifying the requirements in terms of cumulative and peak bandwidth within the latency constraints imposed by inter-city distances in Australia. Distributed computing resources at several centres around the country will be used to demonstrate the need for reserved bandwidth and investigate how it can best be managed. A mix of short term and longer term demonstrations are proposed.
It will be important not only to characterise the bandwidth requirements of particular applications, but also to investigate the network booking patterns associated with typical applications and user communities. It has already been observed, that the ATM bandwidth availability could be much better utilised if an automated booking mechanisms or switched virtual circuit mechanism were provided. Manual booking is not practical for growing user communities, and is wasteful of scarce network resources.
Project objectives are:
These can be met by a mix of focussed short term demonstrations, specially formulated to exercise the network components of distributed systems, as well as some longer term applications activities.
This project aims to address the following demonstrable applications:
Craig Patten will be anchor for the Ninf demo, and Heath James will be anchor for the DISCWorld demo.
Heath James is the anchor for the POVRAY Ray tracing demo with help from Duncan Grove, and Heath is also anchor for the streaming MPEG demo.
Paul Coddington is the anchor for these.
Ben Evans at the ANUSF is the anchor person with responsibility for this work.
These demonstrations for part of a broader portfolio of network demonstratable applications being used by the RDN and ACSys CRCs to study other networks such as the Australian/Japan link and links with the USA. The areas indicated will also complement video-conferencing and protocol investigations to be carried out over the same networks.
In addition to the demonstrations above, an effort will also be made to make straightforward but comprehensive measurements of effective bulk bandwidth and latency using ftp or a network benchmark program. Jesudas Mathew will be anchor for this, with input from the rest of the team.
Longer term demonstrations may be carried out to facilitate access to remote HPC equipment.
The project will require set up of network connectivity between the various sites involved. For each demonstration the following tasks are involved:
In most cases this will involve resource booking although a large part of the work should be possible using compute/storage resources already available in dedicated mode to the project team.
Where possible the measurements will be made over several different networks to ascertain aggregate behaviour.
A detailed management plan will be progressed through email with the project team members and this will be tracked over a project Web page. Responsibilities for the various applications areas are already identified above.
The work will be carried out over a period of approximately 6 months once the AARNet2 infrastructure is in place. It is hoped this will be during the period April-September.
The initial demonstration of connectivity between Adelaide and Canberra needs to be in place for early May, with at least preliminary bandwidth and latency measurements carried out. Melbourne should be involved as soon as is practical, and the applications demonstrations ready to run across all available network configurations when they become available.
It is intended to have the preliminary suite of applications demonstrators ready by end-April.
The project team are drawn from the staff and postgraduate students of the Research Data Networks and Advanced Computational Systems Cooperative Research Centres based at Adelaide and Canberra, the ANU Supercomputer Facility and South Australian Parallel Computing Centre, as well as collaborating colleagues at Melbourne and the relevant Network Support personnel at Adelaide, Canberra and Melbourne.
Special responsibilities for application demonstration components lies with:
Connectivity Team at Adelaide for setting up network connections to the Adelaide CS Dept:
Connectivity team for ANU and ACSys/RDN CRC in Canberra:
Connectivity team for Melbourne University and CSIRO/Bureau of Meteorology - TBA:
10 - 20 Mbits per second connectivity is required between the Adelaide University Computer Science department communications room and:
Raw ATM access is preferred since this would facilitate other complementary work to be carried out in ATM proprietary pseudo-SVC protocols to be carried out as well as investigation of video conferencing performance over the same infrastructure.
The actual applications demonstrations in Distributed and High-Performance Computing will use TCP/IP however. This can be set up on top of ATM.
An automated booking system is highly desirable, and manual circuit setting up each week is unlikely to be useful.
It is desired that the dedicated bandwidth be made available for a minimum of 1 day a week (9am -5pm) between the periods April to September 1998. If it is possible to have this connectivity for the remainder of 1998, further work would be possible.
It is unclear whether the different ATM switch types available at Adelaide, Melbourne and Canberra will present any difficulties. Adelaide has a Fore switch as its primary network connection, with DEC Gigaswitch internally.
The network will need to interface to existing equipment at the proposed sites. Details of this will be worked through in close cooperation with the site RNO representatives.
Ken Hawick Department of Computer Science The University of Adelaide Adelaide SA 5005 Tel: +61 8 8303 4519 Fax: +61 8 8303 4366 Email: khawick@cs.adelaide.edu.au
Others to be arranged.