JSC Coordinates Maestro Project

The transport of data through modern and future HPC architectures is increasingly becoming a key bottleneck. The software stacks currently in use still lack data awareness. This becomes more crucial as memory and storage hierarchies become more complex. This lacking memory awareness, as well as the lacking data awareness, is at the focus of the Maestro project. Together with partners from five European countries, JSC started the three-year EU-funded Maestro project with the goal of providing consistent data semantics to multiple layers of the stack, including both systems software and applications.

The key idea is to introduce a data model where data objects are annotated and the semantics of the data is communicated. Expanding this concept to multiple layers of the HPC software stack enables middleware to handle data transport and manage data transformations transparently to the user at different hardware layers. This will allow application developers to remain insulated from hardware details despite the latter becoming more complex.

The project has adopted a co-design methodology, i.e. applications that are in need of high-performance compute and storage architectures will guide the development work and will be used to demonstrate the benefits and usability of the developed technology. Application areas range from numerical weather prediction, climate and earth system modelling, and materials science to computational fluid dynamics.

The project consortium comprises two leading industrial providers of HPC solutions, namely Cray and Seagate, several leading supercomputing centres, namely CEA, CSCS, and JSC, the European Centre for Medium-Range Weather Forecasts (ECMWF), and the SME Appentra. For further information, visit https://www.maestro-data.eu.

Contact: Prof. Dirk Pleiter, d.pleiter@fz-juelich.de

from JSC News No. 261, 24 October 2018