UK Met Office
EPCC's long-running collaboration with the UK Met Office focused on the development of a parallel version of their Unified Model weather forecasting and climate modelling code.
A range of message-passing systems were used at different stages of the project, from CHIMP through PVM to the MPI standard. The key objectives over the ten-year collaboration included:
- The design and implementation of a parallel scheme for the key Data Assimilation phase of the forecast model. Real measurements of environmental conditions are made by weather balloons, surface ships, ground stations etc and must be assimilated into the computational model to provide an initial state from which to project a forecast.
- The analysis of the feasibility of applying parallelism to the key Unified Model forecasting program and the identification of key areas within the code.
- The design of an overall strategy for parallelisation of the Unified Model.
- The design and implementation of parallel versions of key aspects of the UM code.
- Support in setting up and preparing the parallel Unified Model for competitive benchmarking for an in-house system for the UKMO.
Key achievements included:
- A successful implementation of a parallel Data Assimilation scheme based on EPCC's PUL-TF task-farm library.
- Parallelisation of the core Unified Model program using a three-dimensional regular domain decomposition scheme. Original development was performed with EPCC's CHIMP message-passing system, with later work to port the code to the emerging PVM and later MPI standards.
- Parallelisation of additional key areas of the model code.
- The first successful runs of parallel version of the Unified Model on Meiko CS-2 and Cray T3D systems.
- Additional enhancements to the parallel code to provide bit-identical results with the original serial code.
- Key technology transfer between EPCC and UKMO, enabling UKMO staff to undertake parallel development alongside EPCC.
A long-term strategy for the Met Office is to increase their capacity to run large numerical forecast and climate models. Scalable parallel computing was thus a natural avenue for them to explore and EPCC was pleased to have assisted in formulating their parallel computing strategy. In 1996 the Met Office took delivery of a Cray T3E massively-parallel computer as their main numerical modelling system.