Atos has been selected by the State Meteorological Agency (AEMET) to supply and install your computing and storage technology. Based on Atos’ BullSequana supercomputing architecture, the new supercomputer provides almost ten times more computing capacity than the current one, installed in 2014.
The installation has consisted of the start-up of two clusters equipped with more than 50,000 cores based on AMD EPYC processor technology, with a capacity of 5.9 PB of raw storage, connected by a low latency 200 Gbits Infiniband network / sec
This new supercomputer reinforces and expands the current computing capabilities of the AEMET to promote research in different areas such as weather forecasting, climate change, wave prediction and support its collaboration with international organizations. About the possibilities of this team and the support of ATOS, we have been able to talk with Raul Hilara Parra, telematics coordinator at AEMET.
[MCPRO] How has weather forecasting and climatology analysis changed thanks to supercomputing?
[Raúl Hilara Parra] Supercomputing systems are an essential part in improving forecasts, their common element being, both for meteorology and climatology, the handling of very high volumes of data, and the use of complex probabilistic models, whose efficient treatment determines the accuracy of predictions, and their probability of error.
Thanks to supercomputing, the algorithms that can be solved in reasonable times are increasingly complex, incorporate a greater number of data sources, can simultaneously take into account more boundary conditions, and can even combine several models. The consequence is greater precision, resolution and scope for predictions, even reducing the time needed to obtain results.
[MCPRO] What were the main limitations that they wanted to overcome when launching the new tender? What needs did you want to cover?
[Raúl Hilara Parra] The preparation of a tender for such a specific system has involved a lot of work during the preparation of the specifications. To determine the scope, we identified three strategic lines of action, of which the ‘Improvement of essential public services’ and the ‘Promotion of scientific-technological innovation’ entail the exploitation of state-of-the-art computing resources, with a view to providing AEMET with of the operational capacity for the application of science to the services required by its statute.
Among the needs we had for the new equipment, we needed to improve the ensemble prediction model (2.5 km in the Balearic Peninsula and 1.3 km for the Canary Islands), incorporate a nowcasting system, a dynamic downscaling system for predictions climatological and improve the current resolutions of the models.
Once these requirements were known, they had to be complemented with the considerations of different AEMET operating units to guarantee the correct operation of the machine, such as administration, 24×7 support, training, etc.
[MCPRO] What characteristics are the most valued of the Atos offer?
[Raúl Hilara Parra] The evaluation of the bids was divided into two phases: first, the implementation, operation and support plans of the bidders were evaluated; and secondly, different technical and economic criteria were assessed.
The implementation, operation and support plans of Caixabank-Atos were valued very positively, especially highlighting the proposed test plans; permanent supervision of service elements; and the proposed support plan. Regarding the technical and economic criteria, the results obtained by the test benches (benchmarks) indicated by AEMET to compare the performance of the different proposals of the bidders, and the budget, were valued above all.
[MCPRO] In which areas of action or tasks is the new supercomputer going to be most relevant?
[Raúl Hilara Parra] Technological evolution and the improvement of energy efficiency have meant a notable improvement in performance that translates into new possibilities for both meteorology and climatology.
Among others, we can solve the models in less time, we improve the scope of the predictions (how many hours are predicted in advance), we use more input data in the models, we improve the geographic resolution, we incorporate new applications and we can make predictions of nowcasting, that is, a lot of data can be analyzed in a short time.
In addition, new meteorological and climatological models will gradually be incorporated, such as the gSREPS model, which is a multi-model multi-contour boundary, which will help predict adverse and/or extreme meteorological phenomena, such as heavy rainfall in the Mediterranean area (the famous cold drop), and that they will have a high impact.
