This news blog provides news about the e-IRG and related e-Infrastructure topics.

Back

European exascale projects seek to make programme environments more application suitable

During the Workshops Day at ISC'17 in Frankfurt, Germany,  we were able to talk to Guy Lonsdale from Scapos at the occasion of the ETP4HPC Workshop. Scapos is a spin-off company from Fraunhofer. This SME company was set up to do marketing and sales of high tech products from Fraunhofer but Scapos also works with other SMEs and multiple Fraunhofer institutes. The company has some products in the HPC arena such as highly scalable algorithm multigrid, linear solvers, support for multi-physics co-simulation, and programming layers like GPI, which is a GASPI implementation for PGAS or the LAMA library which supports easy programming across heterogeneous systems. So, it takes a lot of the details away from the applications.

In parallel to that, Scapos has also some activities in the management of European Union projects. The company was involved in the Fortissimo I & II projects, looking at bringing HPC access and use for modelling, simulation and data analytics to manufacturing and industrial settings. Scapos is also managing ExaNoDe which is in the FET HPC programme. This project is developing component technologies for the next generation of the future exascale computing node. On one side, Scapos has HPC relevant products and on the other side, it is focusing on management in the EU programme.

In addition, Guy Lonsdale is involved in the ETP4HPC activities. In particular, he is one of the co-chairs of the working group for programming environments. Together with his colleagues, he is working on the Strategic Research Agenda for HPC towards exascale computing and a particular interest is programming.

The ExaNoDe project is ongoing. It is now in month 18 and is part of a group of projects including ExaNoDe, ExaNeSt and ECOSCALE, looking at different aspects. ExaNoDe is developing the compute node technologies. ExaNeSt is going beyond that to deliver the interconnect, the cooling system and other system aspects. ECOSCALE is looking at accelerators, in particular with FPGA evaluation. These are three projects in a bigger programme of technology development projects for the future European exascale system. It should lead to the themes that are being discussed at the ETP4HPC Workshop, targeting extreme-scale demonstrators. The hope is that these component technologies with the various specific areas should come together in the programme and lead to full platform development and demonstrators towards exascale.

Guy Lonsdale is expecting a lot of political pressure to develop and deploy exascale systems in Europe. The intention is that at least one of these systems is based on European technology. The hope for ETP4HPC is that the programme and the research agenda should develop components that lead and support that political design.

As for the programming environments, this is developing on from the previous strategic research agenda which is being realized partly in ongoing projects. Of course, the classical issue is about hybrid programming models, how to really realize and have a standardized and interoperable MPI + X, where X may be PGAS or shared memory. The thrust now is to really make, not just high performance programming environments but high productivity code environments. It is recognized that one needs to do something to make the programmability for HPC systems, and exascale systems in particular, easy for applications. So, there is a target item to take away some of the complexity and leave the systems with more application specific or application suitable programming environments. That is one of the main threads or the desire to have high productivity. That is one of the focal points.     

Something related to this are the programming languages. Developing new programming languages is a possibility. Domain-specific languages are a theme. With that approach, you could really target specific sectors and provide something that is appropriate for their needs but putting a high focus on developing yet another new language is not part of the projects. However, developing domain-specific languages is included in the research target.