Met Office: Computers

	Computers

From the time of the first computers the Met Office has used state-of-the-art supercomputers for numerical weather prediction and, more recently, also for predictions of global climate. Developments in computer hardware technology, combined with advances in numerical methods, and an increased understanding of atmospheric and oceanic processes over this period, have lead to steady improvements in the accuracy and timeliness of forecasts.

The Met Office deployed computer systems over this period reflect evolving trends in supercomputer technology from the scalar IBM 360/195, through the vector Cyber 205, the multi vector-processor Cray C90, massively parallel Cray T3E and the current NEC SX-8/SX-6. The Met Office has consistently used state-of-the-art computer architectures for achieving near-optimal performance from its numerical models.

More about the history of computers at the Met Office

NEC SX-8/SX-6

NEC SX-8 supercomputer

In early 2005 the Met Office accepted the first production NEC SX-8 supercomputer in the world. This system provides additional capacity to the NEC SX-6 computer systems used by the Met Office in Exeter. The SX-6 and SX-8 machines are divided into 'nodes' with each node containing eight processors. Currently the SX-8 has 25 nodes, and there are two SX-6 clusters; one with 19 nodes, and the other with 15 nodes. Each SX-8 node is twice as powerful as an SX-6 node, so together, the combined systems deliver over thirteen times the sustained power of the previous Cray T3E computers, which the SX-6 replaced in 2004.

For operational resilience, the systems are divided between two halls, with the SX-8 and the 15-node SX-6 cluster in one hall, and the 19 node SX-6 cluster in the other. This enables the operational forecast to be maintained even if one of the halls is unavailable for any reason.

The nodes are accessed through front-end machines. There are six of these, three in each hall. The front-end machines are scalar machines and perform tasks such as compiling computer code, which is better performed on a scalar machine. The resulting program is then run on the vector nodes.

As was the case on the previous supercomputer, a Cray T3E, the operational model is run on multiple processors but as each individual processor is that much more powerful than the T3E ones, far fewer are required. For instance, the operational global model is typically run on 4 nodes compared to 144 T3E processors.

NEC SX-6 supercomputer