Presentation on theme: "Educational Issues in E- Science in Australia Kevin Burrage Federation Fellow of the ARC – UQ, Brisbane, Australia October 2004."— Presentation transcript:
Educational Issues in E- Science in Australia Kevin Burrage Federation Fellow of the ARC – UQ, Brisbane, Australia October 2004
Contents n Background: u Education in Australia u APAC u APAC Partners u GRANGENET u QPSF n Two Exemplar Projects in QPSF: u Teaching via Access Nodes u Multimatlab n Issues and Conclusions.
Education in Australia n 35 Universities n 19 Million people n Vast distances n Good communication networks n Higher Education – a multibillion dollar industry n Many overseas graduate students (Asia) n Overseas campuses – expensive n Initiatives such as Universitas 21.
Australian Partnership in Advanced Computing n n Research support. n Education. n Tech Diffusion. Education n n Shared educational modules between universities at graduate and undergraduate levels.
APAC Partners and Projects (Education) Modules in n AC3 – SMP programming and VTK. n ANU – graphics, data mining, MPI. n IVEC – bioinformatics – web servers, graphics. n QPSF - computational engineering, Multimatlab and teaching via access nodes. n SAPAC – MPI. n TPAC – visualisation for earth systems, digital libraries. n VPAC – engineering case studies.
GRid And Next GEneration Network n n Three year program to operate a multi-gigabit network supporting IPv6 and IPv4 and multicast. n 10 gigabit backbone - Melbourne,Canberra,Sydney n 5 gigabits into Brisbane. n Small annual membership fee for the three years. n Used for all Research and Education (R&E) traffic.
QPSF n Consortium of 6 Queensland Universities. n All with access nodes. n n James Cook University - Masters in Computational Engineering with the University of Wales. n Griffith University - undergraduate courses in HPC. n The University of Queensland - courses in HPC, Scientific Visualisation, computer graphics. Access nodes in treaching. New Major, within BSc. degree, in Computational Science. n QUT - Graduate Diploma and Masters in Scientific Computation and Visualisation.
Graduate course in Graphics via AccessGrid Sessions n Semester 2, 2004: UQ and UWA shared 6 lectures for a component of an Advanced Visualisation course. u 10 students at each institution. u Dr Pamela Burrage (UQ) - Curves and Surfaces. u Dr Karen Haines (UWA) Programmable Graphics PU. Outcomes: n University credit for students attending the subject. n Students found the experience interesting and valued the chance to attend lectures given by leaders located on the other side of Australia. n Students enthusiastic about attending the lectures.
Other Issues n Lecturers need to consider the presentation style, to encourage interaction between students (at each venue) and the lecturer. n Need a whiteboard, for impromptu explanations or diagrams of course content. n Need a way of running programs easily at each venue. n There were some teething problems: u The audio in one session was patchy. u Used Shared Presentation using OpenOffice but not all equation objects (in powerpoint) were displayed properly. u Advantage of OpenOffice is that only 1 person needs to control the advance to the next slide.
MultiMATLAB Background n Cleve Moler - mid-1980s; an Intel iPSC. ( Why there isnt a parallel MATLAB, MathWorks Newsletter, 1995). n Anne Trefethen ; MATLAB was run on multiple nodes of an IBM SP-1, using a Fortran wrapper. n MathWorks support the initial development of MultiMATLAB by a group of researchers at the Cornell Theory Center (A. Trefethen, C. Myers) and the Department of Computer Science (V. Menon, C. Chang, G. Czajkowski, N.Trefethen). n MultiMATLAB project was dormant before being revived by Anne Trefethen (on a 1-year visit to ACMC-University of Queensland) and G. Ericksson (ACMC) in 2004.
Aim n to use MultiMATLAB as a teaching tool for parallel programming in a third year computational science subject at UQ in Semester 1, Issues n Students in scientific computing are doing less programming in Fortran and C. n Students already familiar with MATLAB will be able to learn parallel programming in a known environment. n This will allow them to focus on actual parallel programming constructs which they can implement using a few simple MultiMATLAB commands.
Background n MultiMATLAB allows multiple instances of MATLAB to communicate with each other, with MATLAB scripts parallellised across multiple processors. n A user starts up multiple instances of MATLAB These processes are usually run on separate processors. They can execute commands on the other processors (e.g. Eval(…);) or distribute the data amongst the processors (e.g. Distribute(x);) n The usual message-passing commands are available (e.g. Send, Recv, Bcast) as well as global reduction commands (e.g. Min, Max, Sum). n MultiMATLAB provides multi-processor Graphics.
Conclusions and Issues
Access Grids in Education n Get the right supporting technologies. n Small scale subjects only. n Fee paying courses into Asia – Universitas 21. n Ideal for sharing material and lecturer expertise between universities – smaller universities. n Need to broaden user base through small grants. Other issues n Repositories. n New paradigms for teaching HPC – Multimatlab.
Thanks n Pamela Burrage, Geoff Ericksson, Steve Jeffrey (UQ) n Karen Haines (UWA) n Anne Trefethen (UK E-science)