After Control Data Corp decided to liquidate its ETA Systems supercomputer business, it turned to Cray Research Inc and asked the company to take care of its customers, and the way the agreement is intended to work is demonstrated in the arrangements being made for the Meteorological Office in the UK (CI No 1,229). The four-processor ETA 10 system installed at the Meteorological Office’s Bracknell headquarters last June has not been working to spec and is to be removed from the site as soon as is practical. Control Data is to supply a comparable eight-processor Cray Y-MP-8/32 system in its place, and this will be delivered early next year. In the meantime, weather forecasting will continue on the existing Control Data Cyber 205, which is to receive a UKP1m upgrade, including a quadrupling of its memory and a doubling of the vector pipe facility, extending its useful life at the Met Office until 1991 when the Cray machine should come into service. The Met Office will have to pay out for the upgrade, but the more expensive Cray Y-MP is to be supplied at no extra cost under Control Data’s agreement with Cray. The weathermen reckon the upgrade should boost the Cyber’s power by about 75%, and while it is being carried out, forecasting will run on one of the four processors on the ETA 10 machine – said to be comparable with the performance of the Cyber machine. The system is currently front-ended by an IBM mainframe, as most of the supercomputer’s workload is concerned with the batch processing of data from the many collection points, but there will be a move to Unix workstations over time. The changes mean that there will be delays in the Met Office’s plans for new forecasting models – the Cray will be its first Unix venture – though the process is to be accelerated as much as possible. The new model is to offer improved forecasting, with data collection being performed at a 45 mile resolution doubling the present 90 mile model – which requires a tenfold increase in computing power and the need for much better communications and portability standards. The objective is to integrate global long range forecasting with local day-to-day weather predictions, as well as offering very short period forecasting of only a few hours ahead, which will incorporate recommendations from the report on the failure to forecast the hurricane that devastated the South of England back in October 1987. In the meantime, Michael Fish will just have to keep his fingers crossed and his seaweed dry and pray that the old Cyber doesn’t miss the next hurricane.
