Building the exascale future

Author: Michele Weiland
Posted: 20 Nov 2015 | 15:15

We are very proud to be leading two new projects focused on the Exascale challenge: NEXTGenIO and INTERTWinE. Both started on October 1st 2015 and will run for 36 months.

The European Commission has recently funded a round of Horizon 2020 FETHPC research projects focused on the Exascale challenge. In total, 19 new projects successfully bid for this funding: 6 of them, the ‘larger’ projects, investigate Exascale system solutions, whereas the 13 ‘smaller’ projects address more specific problems, such as algorithms or APIs.


NEXTGenIO is one of the larger projects. It addresses a key challenge not just for Exascale, but for HPC and data intensive computing in general: the challenge of I/O performance. As core-counts have increased over the years, the performance of I/O subsystems have struggled to keep up with computational performance and have become a key bottleneck on today’s largest systems. 

NEXTGenIO will develop a prototype computing platform that uses on-node non-volatile memory, bridging the latency gap between DRAM and disk, thus removing this bottleneck. In addition to the hardware that will be built as part of the project, NEXTGenIO will develop the software stack (from OS and runtime support, to programming models and tools) that goes hand-in-hand with this new hardware architecture. Two particular focal points are a data and power-aware job scheduling system, as well as an I/O workload and workflow simulator that will allow us to stress-test our developments.

We believe that the new platform being developed by NEXTGenIO will be capable of delivering transformational performance. 


INTERTWinE addresses a problem that is already apparent today, and will become increasingly critical as heterogeneous and hierarchical system architectures become more complex. It is unlikely that in the short to medium term there will be a single programming model or API that can target all aspects of such an architecture effectively. INTERTWinE therefore tackles the interoperability of parallel programming models that each target specific levels of the system hierarchy, both on an implementation and specification level.