The UK strategy is to always have one of the fastest supercomputers in the world avaialble for scientists. When technology advances, scientists should get access to a new system as soon as possible.

Currently there are several scientific supercomputers installed in the UK. A consortium of the University of Edinburgh, Daresbury Laboratory, CCLRC and IBM, installed the HPCx as it is called: a 3.4 Tflop/s system, that should be expanded to 6 Tflop/s next year and 12 Tflop/s in 2006. CSAR in Manchester will install, later this year, an SGI Altix 3000 with 1 Tflop/s performance.

For these systems, there is Computational Science and Engineering (CSE) Support of Daresbury Laboratory.

The current strategy for UK HPC services, Pilcher-Clayton explained, is to initiate a competitive procurement exercise every 3 years for both the system and service. After successful procurement a contract will be awarded for a service with a 6 year life span. The current one is called HPCx, the next one HPC(Y), and the one after that HPC(Z). At any one time, 2 services will overlap by 3 years.

Allthough this is a sound strategy from the viewpoint of continutity, there are also some issues left. The user community of the UK national services is relatively small. The procurement process for the national services emphasises competition rather than collaboration amongst the players.

The life expectancy of 6 years for the national services does not fit with the 'natural' timescales of the separate elements of the support which is larger than 6 years and system technology developments which go faster than 6 years.

Pilcher-Clayton complained that the costs of staying at the forefront of the technology are escalating. It is also costly to have different systems with different technologies to finance, that require different skills for maintenance and application support.

The procurement of HPC(Y) calls for an initial service capability of a peak performance of 50-100 Tflop/s in 2005, doubling to 100-200 after 2 years and doubling again to 200-400 2 years after that.

They are looking for more components of the service than just one hardware system. This should also contain large capacity data management, visualisation facilities, CSE support, and training.

In the procurement process, several lessons learnt from the procurement of HPCx will be taken into account, Pilcher-Clayton said. The procurement process has delivered a solution which promises excellent value for money but it takes too long. It is important to consider when to enter the market and the process of understanding and capturing user requirements needs to be strengthened. It is, according to Pilcher-Clayton, crucial to carry out a full market survey before the finalisation of requirements.

He closed with presenting the time-table:

• October 2003, development of a scientific case, using a Trends and Opportunities Panel, and Technology Watch Panel.
• October 2003, Lancaster University to deliver a methodology for the capture of HPC requirements;
• November 2003, hold one-to-one meetings with the vendors at SC2003;
• November 2003 to March 2004, capture requirements (including codes for benchmarking) and develop the Statement of Requirements;
• April 2004, issue the OJEC notice;
• December 2005, start the service.