A Scalable Object Store for Meteorological and Climate Data

Overview of high-performance data infrastructure at ECMWF

In this video from PASC17, Simon D. Smart (ECMWF, United Kingdom) presents: A Scalable Object Store for Meteorological and Climate Data.

“Numerical Weather Prediction (NWP) and Climate simulations sit in the intersection between classically understood High Performance Computing (HPC) and the Big Data / High Performance Data Analytics (HPDA) communities. Driven by ever more ambitious scientific goals, both the size and number of output data elements generated as part of NWP operations has grown by several orders of magnitude, and will continue to grow into the future. The total amount of data is expected to grow exponentially with time, and over the last 30 years this increase has been approximately 40% per year. This poses significant scalability challenges for the data processing pipeline, and the movement of data through and between stages is one of the most significant factors in this. At ECMWF, meteorological data within the HPC facility is stored in an indexed data store for retrieval according to a well defined schema of meteorological metadata.”

This paper discusses the design and implementation of the next version (5th) of this indexed data store, which aims to increase the range of contexts within the operational workflow in which it can be used, and to increase its tolerance to failure. Further, it aims to pre-emptively head off some upcoming scalability bottlenecks present in the previous versions.