A unified measurement model for meteorological computing resources

High performance computer (HPC) capability building has made raped progress in national headquarter and regional centers of China Meteorological Administration. To resolve fine-grained resource allocation, distributed resource sharing and scheduling, and system introduction evaluation, a unified measurement model for meteorological computing resources with wide applicability and practicability must be established. Taking into consideration related research and existing methods, a general, hybrid measurement model is proposed, integrating both the conventional industry benchmarks and meteorological domain-specific applications. As the core of the model, GCUP (General Computing Unit Productivity) is introduced as a common form to implement quantitative representation of heterogeneous computing resources. An implementation of the model has been made, including 5 object parameters. Performance benchmark and meteorological numerical model tests are run on 4 representative HPC systems distributed in national headquarter and regional centers of China Meteorological Administration. Based on GCUP data determined, an initial nationwide HPC resource usage statistical analysis is made.