Following its acquisition of the first university-based supercomputer in Australia in 1987, the University undertook to provide ten per cent of the capacity of the vector supercomputers free of charge to researchers at other Australian universities. Over the ensuing decade, the External Use Scheme has proven very successful and in 1997 it was extended to provide access to the Silicon Graphics PowerChallenge system in accord with the policy of the Australian Cooperative Supercomputer Facility.
Until 1993, the ANU did not involve itself in deciding the relative merits of projects from external universities. Over the first five years of this External Use Scheme, a formula provided by the Australian Vice-Chancellors' Committee was used to allocate time in bulk to each participating university. By the end of 1992, it was clear that the system needed modification. The number of projects grew to be too large, with the result that each project received an inadequate grant. Therefore, after discussions with a number of external coordinators and users, the scheme was altered so that the Supercomputer Time Allocation Committee accepts proposals from external researchers for a reserved ten percent of the computer time available separate from ANU allocations. In considering proposals, the Committee consults national and international referees and is guided by the same principles of academic merit and suitability of the resources to the proposed project as it uses for internal proposals.
In 1997, 45 external projects were supported on the VPP and 13 on the SGI-PowerChallenge (4 projects had access to both systems). The VPP system in particular was heavily oversubscribed. Nineteen of these projects had independent support from the Australian Research Council.
Tables 1 and 2 list the external projects which were supported. The 45 VPP projects involve 104 researchers and the 13 SGI-PowerChallenge projects 33 researchers. Omitted from the table are those projects which were given small provisional grants and which made little or no use of the grant. Detailed descriptions of the projects can be found in Appendix B.
In 1997, the University of Queensland joined the Australian Cooperative Supercomputer Facility. This collaborative venture was initiated by the ANU in 1991 in conjunction with regional computational consortia representing almost all the New South Wales and South Australian universities. The ACSF operates with the support of several substantial Australian Research Council infrastructure grants and was successful in obtaining additional RIEF funding for 1998 from the ARC in 1997.
The ACSF members operate large Silicon Graphics PowerChallenge systems with installations at the ANU, NSW (Australian Technology Park), the University of Adelaide and the University of Queensland.
In addition, ACSF members now share a Connection Machine CM5 operated by the South Australian Centre for Parallel Computing. In 1997, the ANU shipped its CM5 to Adelaide, enabling our colleagues in Adelaide to build a substantial 128 mode system (16 Gflops) from the systems formerly owned by the ANU, NSW and SA. When built, this system was the second most powerful computer available in Australia.
The ACSF has undertaken to implement an external use scheme for researchers from other universities requiring access (see previous section). Researchers at universities within the ACT have access to the Silicon Graphics PowerChallenge system at the ANU as an ACT regional node, i.e. their access is on the same basis as ANU users and is not confined to the ten percent reserved for other external universities. Researchers requiring access to the SGI-PowerChallenge from universities in regions which have another ACSF node (SA, NSW and QLD) are generally referred to their regional node, although access is provided for those requiring software or other environmental conditions only available at the ANU.
Late in the year the ACSF partners were joined by the University of Western Australia in making a proposal for a national Centre of Expertise in High Performance Computing and Communications to DEETYA.
Dr Gingold attended a number of ACSF board meetings. Facility staff have also been active in the SGI/Cray HPC User Group which is directly relevant to ACSF, attending two interstate meetings and organizing one at the ANU during the year.
The RASH resource and system control software developed and marketed by the Facility for both the Silicon Graphics and Fujitsu systems is available free of charge to ACSF members. It is currently installed on the NSW and ANU systems.
In previous years workshops specifically for external postgraduate students have been conducted. While this did not occur in 1997, a substantial number of external users, mostly postgraduate students, attended the training courses listed in this report under Education and Training.
A research agreement between the ANU and the Galician State in Spain was signed by the ANU and the Galicia in 1997. This followed visits to the Centro de Supercomputacion de Galicia (CESGA) in Santiago de Compostela, Spain by Dr Gingold and Professor Stanton, DCS and talks with Mr Celso Curras, the Minister for Education of Galicia. The agreement makes funds available to assist in developing collaborative projects in specific fields.
A number of visits of ANU researchers to Japan and vice versa have been funded over the years under the research and development agreement with Fujitsu Limited entered into in 1987. In 1997 the fund was used to support the attendance of Dr Henry Gardner, DCS/ANUSF to attend both the Second Asia Pacific Plasma Theory Conference, Japan from 24-26 September and the Eighth International Toki Conference from 29 September - 3 October. At the latter he presented three papers related to computational studies of the H-1 National Facility Heliac fusion reactor carried out on the VPP300 with his students and colleagues in RSPhysSE. He also discussed a proposal for a cooperative project between the ANU and the Institute of Advanced Energy at Kyoto University to optimise software for modelling stellarator fusion reactors on VPP architectures.
The Area 3 collaborative software development project with Fujitsu Japan was established in 1989 with the aim of porting and optimizing computational chemistry software for the VP-100. Over the years the scope of the project has varied and included elements of mathematical and climate modelling software (in conjunction with DCS and CRES) as well as the preparation of several scientific videos. Recently the project has been refocussed on the porting and optimizing of chemistry software and its application in a research project with JCSMR. Work now spans both the vector parallel (VPP) and scalar parallel (AP) range of Fujitsu supercomputers.
The collaborative research project with Dr J Gready of the Computational Molecular Biology and Drug Design Group, JCSMR continued in 1997. The aim of this project is to take the computational chemistry software ported to and optimized for Fujitsu parallel systems, and apply it to the study of enzyme reaction mechanisms. Specifically the size of the problems to be tackled through this joint initiative are at the cutting edge of what is currently feasible, and are only possible by careful optimisation, parallelisation and extension of the software used.
The core software packages of interest to the project in 1997 include the GAUSSIAN 94 and GAMESS-US quantum chemistry codes, the MOPAC and MNDO94 semiempirical codes, and the AMBER and X-PLOR molecular dynamics/mechanics codes. The research project has been using these codes for the study of the enzyme Dihydrofolate Reductase (DHFR). This enzyme is important in processes that construct cellular DNA, and inhibition of the mechanism has been explored by several classes of cytotoxic drugs for treating diseases such as cancer. DHFR catalyses the reduction of a folate substrate to dihydrofolate and then to tetrahydrofolate in the presence of an NADPH cofactor (figure available if desired). Although this reaction has been studied both experimentally and theoretically a number of times before, several aspects of the mechanism remain unclear, and led us during the last year to study the polarisation of the substrate and cofactor in the active site. This work has involved performing quantum chemical calculations on systems containing up to 124 atoms and using close to 2000 basis functions.
During 1997 the project applied for and was awarded an ARC sponsored Australian Postgraduate Award Industry (APAI) with Fujitsu as the Industrial sponsor. The proposed project seeks to investigate the use of MOZYME (a linear scaling version of MOPAC that has been developed by Dr J.J.P Stewart with the direct support of Fujitsu) for the study of enzyme reaction mechanisms, and specifically to contrast the use of MOZYME with existing hybrid quantum chemical/molecular mechanical (QM/MM) methods. The PhD student supported by this grant will be located in JCSMR and is expected to start in early 1998. Work on this project will be carried out in collaboration with Dr Nobes, formerly leader of the project in ANUSF and now at the Fujitsu European Centre for Information Technology (FECIT), since he is responsible for the porting and parallelisation of MOZYME for the VPP and AP platforms. In addition we are interacting closely Dr Stewart, the author of MOZYME, and were please to have him visit the ANU as part of the Area 3 project in September 1997.
The project is led by Dr Rendell, ANUSF who is partially supported by Area 3 funds and is responsible for the day-to-day management and planning. The Area 3 project supports 3 full-time staff working in ANUSF, a postdoctoral fellow in JCSMR and the soon to be appointed APAI PhD student. During the year there were many contacts between the ANU and Fujitsu Japan, including two project planning meetings attended by Dr Rendell in Japan. As well as direct contact with Fujitsu Japan a close collaboration exists between the Area 3 work being performed at the ANU and complimentary work undertaken by FECIT in the UK. In 1997 this lead to Dr Rendell visiting FECIT and presenting seminars at FECIT and at a number of other Fujitsu sites in Europe.
Area 3 Publications
A. Bliznyuk, A. C. Limaye and A. P. Rendell, Does Molecular Modelling still need Supercomputers?, Third Australian Molecular Modelling Workshop, Melbourne, 5th - 7th October 1997.
A. Bliznyuk and J. E. Gready, A new VDW-FFT Method for Docking Small Ligands to Proteins, Third Australian Molecular Modelling Workshop, Melbourne, 5th - 7th October 1997.
A. Bliznyuk, A. C. Limaye and A. P. Rendell, Computational Chemistry on Fujitsu Parallel Supercomputers, Parallel Computing Workshop PCW97, 25-26 September. 1997.
S. P. Greatbanks, J. E. Gready, A. C. Limaye, and A. P. Rendell, Computational Studies of Polarisation in the Reaction Mechanism of the Enzyme Dihydrofolate Reductase, American Chemical Society National Meeting, Las Vagas, 7-11 September 1997.
S. P. Greatbanks, J. E. Gready, A. C. Limaye and A. P. Rendell, Computational Studies of Polarisation in the Reaction Mechanism of the Enzyme Dihydrofolate Reductase, Third Australian Molecular Modelling Workshop, Melbourne, 5th - 7th October 1997.
S. R. Gadre, S. S. Pundlik, A. C. Limaye and A. P. Rendell, Electrostatic Investigation of Metal Cation binding to DNA Bases and Base Pairs, Chemical Communications (in press).
During the year there were many contacts between Fujitsu Japan and ANU relating to this project, including visits by Dr Rendell to Japan.
Parallel Mathematical Subroutine Library Project
with Fujitsu Limited - Area 4
The Area 4 project commenced in 1992 and involves research and development of mathematical library algorithms and code for Fujitsu's VPP300/700 supercomputers. The Supercomputer Facility manages this project and three of the staff members are actively involved in the work. Dr M Kahn is responsible for day-to-day management of the project and the coordination of planning and activities across campus. Academic direction and leadership of the project is given by Professor R Brent, CSL, RSISE and Professor M. Osborne, Program in Advanced Computation, CMA, SMS. Staff and students working on the project are located in SMS, Computer Science Laboratory, RSISE and ANUSF.
Three research fellow positions are funded under this project. Throughout 1997 these positions were held by Dr David Harrar II, Dr Lutz Grosz and Dr Zhou Bing Bing. Dr Markus Hegland, CSL, RSISE continued his role in the project, in particular, as supervisor to Mr Geoff Keating who holds an Australian Industry Postgraduate Award scholarship to work towards a PhD in the Computer Science Laboratory, RSISE. This scholarship is supported by Fujitsu.
Within ANUSF Mr Gavin Mercer worked half-time as software engineer and program librarian for the project whilst Dr M Dow and Dr M Kahn were actively involved in algorithm and code development. Dr Dow, who is employed half-time as an Academic Consultant, worked additional time on the Area 4 project and Dr Kahn worked extended hours to manage the project.
Releases of updated versions of the Fujitsu SSLIIVP and SSLIIVPP libraries during the year have brought many more of the Area4 developed routines into commercial usage. The performance of these routines, both for one-processor and multi-processor codes, exceeds that of other mathematical subroutine libraries on the VPP300.
Area 4 Publications
D. L. Harrar II, M. H. Kahn and M. R. Osborne, Parallel Eigenvalue Routines on the Fijitsu VPP300, Parallel Computing Workshop PCW97, P2-Y-1-8, 25-26 September, 1997, ANU, Canberra.
G. Keating, Choosing Trees for FFTs, Proceedings of the Parallel Computing Workshop PCW97,P2-X-1-6, 25-26 September 1997, ANU, Canberra.
O. Nielsen, G. Mercer and M. Hegland, 1997, Vector-Parallel Fast Wavelet Transforms , Proceedings of the Parallel Computing Workshop PCW97, P1-S-2-8, 25-26 September 1997, ANU, Canberra.
M. Osborne and M. Hegland, Wrap-around Partitioning for Block Bidiagonal Linear Systems, Proceedings of the Parallel Computing Workshop PCW97, P2-V-1-5, 25-26 September 1997, ANU, Canberra.
B. Zhou and R. P. Brent, A Parallel Ring Ordering for Efficient One-sided Jacobi SVD Computations, Journal of Parallel and Distributed Computing 42 1-10 (1997).
D. Harrar and M. Kahn, On the Efficient Solution of Symmetric/Hermitian Eigenvalue Problems on Parallel Arrays of Vector Processors, presented at the Computational Techniques and Applications Conference, CTAC97, University of Adelaide, 29 September - 1 October 1997. To appear in the CTAC97 Proceedings, World Scientific.
D. Harrar and M. Osborne, Composite Arnoldi-Newton Methods for Large Nonsymmetric Eigenvalue Problems, presented at the Computational Techniques and Applications Conference, CTAC97, University of Adelaide, 29 September - 1 October 1997. To appear in the CTAC97 Proceedings, World Scientific.
G. Keating, The Six-step Algorithm for Two-dimensional FFTs, presented at theComputational Techniques and Applications Conference, CTAC97, University of Adelaide, 29 September - 1 October 1997. To appear in the CTAC97 Proceedings, World Scientific.
M. Osborne and D. Harrar, Inverse Iteration and Deflation in General Eigenvalue Problem, Mathematics Research Report No. MRR 012-97, Australian National University, submitted.
B. Zhou, The Multiple Minimum Degree Ordering with Greedy Graph Compression for Vector Computers, presented at the Computational Techniques and Applications Conference, CTAC97, University of Adelaide, 29 Sept - 1 Oct 1997. To appear in the CTAC97 Proceedings, World Scientific.
Dr Gingold served on the Duffield Scholarship Committee at MSSSO and is a member of the Global Change Confederation Steering Committee.