The Supercomputer Time Allocation Committee (STAC) has had the responsibility for allocating time on the Fujitsu vector processors since 1987 and on the Connection Machines from 1993 until March 1996. In 1996 the STAC began allocating resources on the SGI-PowerChallenge and the Fujitsu VPP300. Existing grants on the VP2200 were simply extended beyond March until the system was decommissioned.
The Committee's policy is that all but small start-up allocations of time to local researchers should be subject to a peer-review process. Applicants are required to provide an academic justification for the amount of time requested, as well as information by which the suitability of the computer system for the proposed research can be judged. The Chairman of the Committee obtains reports on proposals from referees throughout Australia and in many cases from overseas. New projects and continuing projects that make large requests for resources are routinely refereed. The Committee makes grants of time on the basis of these reports and on the effective use of previous grants. Since the time between meetings of the Committee is quite long, procedures are implemented to assist researchers who wish to commence projects that evolve between meetings of the Committee.
The Committee processed two rounds of proposals during the year. Given the large increase in resources available and the varying power of the VPP and SGI-PC systems as they were upgraded during the year, the process was relatively informal in 1996. The CM5 users continuing to use the system after March were no longer subject to STAC resource allocation.
Members of the Committee are academics appointed on a two-year rotating basis by the Chairman. The Chairman is appointed by the Vice-Chancellor. Professor Evans, RSC replaced Professor Levick as Chair of the STAC in mid-year. Professor Levick had served on the STAC since 1987 and his thorough and unstinting efforts, both as Chairman and member, are much appreciated by the University.
During 1996, members were:
The year was very unusual in that the VP2200 continued operations while the VPP and SGI-PC were installed and progressively upgraded. As a result of this and the difficulties in implementing the accounting software on new systems, the statistics this year are somewhat less reliable and more difficult to interpret than in the past. In particular, not all usage has been accounted for on the VPP300 and SGI-PC.
In the past the total hours available to be granted in a year was around 8000. With the new systems, the total number of hours available has leapt to over 250,000. It will clearly take some time for the users and the STAC to adjust to this substantial change.
Table 3 presents a summary of the number of researchers and time allocations by research field on each system and overall.
A total of 58 internal projects used the VP2200 before it was decommissioned mid-year. This involved 36 Principal Investigators and 128 collaborators. There were only 3 new projects. Usage is reported in Table 4.
A total of 62 internal VPP300 projects were approved by the Time Allocation Committee in 1996. There were 12 new projects not previously active on the VP2200. During the year, there were 128 researchers associated with ANU projects on the VPP300, comprising 37 Principal Investigators and 112 individual user accounts on the system. (The totals obtained by summing the columns headed 'Number of Researchers' in the following tables are greater than 128 because some researchers work on more than one project.)
Grants are allocated in Service Units. A Service Unit is related to cpu use and the priority of the job; at normal priorities, one Service Unit corresponds to one hour of cpu time on a single processor.
Grants to ANU projects amounted to 45,604 Service Units, and the total CPU time used by ANU projects was greater than 34,000 hours. Operational problems led to substantial amounts of lost time but often an amount of work equivalent to a week's work on the VP2200 was processed during a single day's working hours. After September the system became stable and useage levels climbed towards saturation. At the time of writing this Annual Report in early 1997 the VPP usage had become completely saturated.
Note that time was also used by external users, for systems development and for the Fujitsu collaborative software development projects and similar activities.
Table 5 lists grants and usage by project. A detailed description of each project is presented in Appendix A.
A total of 51 projects involving 29 Principal Investigators and 104 researchers were awarded allocations totalling 72,553 Service Units. Total cpu usage was 75,913 hours. There were 16 new projects (9 of which were also new projects on the VP2200/VPP300 projects). There were 39 projects which also were allocated time on the VPP system.
Table 6 lists grants and usage by project. A detailed description of each project is presented in Appendix A.
As outlined in the Collaboration and Outreach section of this annual report, the STAC continued to take responsibility for allocating the bulk of external usage of the VPP300. Details of the outcome and usage are presented in that section.