D
The development of the European HPC ecosystem has to ensure the coordination of technology developers and suppliers, users and procurers. A significant integration and co-design effort is needed to gather the players in the HPC ecosystem and focus on the advancement of technologies, HPC resources, tools and methodologies that respond to the user needs.
The next steps in the European HPC strategy will be the large-scale platform integration of the technology building blocks with a co-design approach, complemented with the coordinated acquisition at EU level of world-class systems in order to transition the European leading supercomputers to the new computing generation, and the further integration of the European HPC infrastructure. This is similar to the approach of the U.S.A. Department of Energy (DoE) for their Exascale Computing Initiative.
evelopment of the European HPC eco-system
The overall objective of the European HPC strategy "ensuring European leadership in the supply and use of HPC systems and services by 2020" can be interpreted as Europe having a HPC ecosystem that is on a par with the best in the world, and that Europe aims to be number one in the world in targeted aspects of HPC.
The HPC Communication outlined a number of measures for the development of the HPC ecosystem, ensuring the coordination of technology developers and suppliers, users and procurers. This requires a synergetic implementation of the three axes of the HPC strategy, ensuring the coordination of technology suppliers and the users and procurers for the definition on exascale requirements, including industrial requirements, and support and accelerate the development of pre-competitive, innovative HPC technologies and systems towards exascale that respond to the user needs. Section 6 above details the concrete efforts supported by the EC in developing this ecosystem.
As indicated in the HPC Communication, an integration and co-design effort is needed to gather the players in the HPC ecosystem and focus on the advancement of technologies, HPC resources, tools and methodologies. Co-design in HPC is necessary as the needs of science and industry will best be served if new hardware and software are designed and developed in close coordination with each other, rather than separately (as has often happened). Technological advances produced in isolation require several years of development of the necessary software and application environment. In a highly competitive environment, this delay is not acceptable.
Several European projects such as Mont-Blanc75 and DEEP76 are making leading-class advances in HPC technologies, using the co-design approach:
Mont-Blanc is designing a new type of computer architecture capable of defining future global HPC standards that will deliver exascale performance while using 15 to 30 time less energy. The reduction of energy consumption will be achieved by developing a full energy-efficient HPC prototype using low-power commercially available embedded technology. Mont-Blanc is a world-leading approach in the use of low-power technologies that can radically change the way supercomputers are built, programmed and use
DEEP - Dynamical Exascale Entry Platform- is developing a novel, exascale-enabling supercomputing platform with lower energy consumption (water cooling with a more than 200x efficiency improvement over comparable air-cooled systems) but with a monitoring system of unprecedented accuracy and granularity relating application activity to energy consumption, which helps users optimize and tune their system according to their needs. DEEP is suitable for a wide range of codes with either limited or high scalability requirements, thanks to its heterogeneous architecture, with a programming model and a dedicated development and runtime environment which makes porting applications easy.
The next steps in the European HPC strategy will be the large-scale platform integration of the technology building blocks with a co-design approach, complemented with the coordinated acquisition at EU level of world-class systems in order to transition the European leading supercomputers to the new computing generation, and the further integration of the European HPC infrastructure. This is similar to the approach of the U.S.A. Department of Energy (DoE) for their Exascale Computing Initiative77 (encompassing the development of building blocks and their integration into co-design platforms, the development of exascale-capable applications and software, and the coordinated acquisition for exascale platforms and systems deployment). In detail, the steps are the following:
-
Technological development of key building blocks (e.g. software stack including programming models, node architectures, algorithms, and a first-level co-design);
-
Integration of the technological building blocks into extreme-scale co-design platforms to first reach pre-exascale performance level by 2019/2020 (0.2-0.3 Exaflops at ~20 MW i.e. 10 times today's performance with the same energy consumption) and to achieve exascale performance by 2021-2022 within the same 20 MW envelope, with the necessary exascale application software for commercial exploitation by 2023. These platforms serve as ‘stepping stones’ towards a more expedited exploitation and better services for scientific and industrial use of the underlying system design and technology. Exascale applications software packages will then be ported enabling monitoring of extreme parallelism, reliability, resiliency and scalability;
-
Coordinated acquisition strategy. In cooperation with PRACE, the platform integration will be complemented with procurement/acquisition of pre- and exascale systems for transitioning the European set of leading machines to the new computing generation (the U.S.A. Department of Energy plans for acquiring pre-exascale systems in fiscal years 2016 through 2022 and an initial exascale system in 2023-2024).
The EU strategy needs to ensure that all stakeholders in an exascale ecosystem can participate in such integration and co-design activities. The EXDCI support action is a first step in this coordination. This will require ETP4HPC to collaborate closely with PRACE, CoEs and other users to ensure that the ETP4HPC roadmap for indigenous technology development reflects the technology and system-level innovations required by end-users and procurers, and to test hardware and software technologies at large scale in the appropriate environments.
The aims are to accelerate HPC innovations by European suppliers and to increase the likelihood that these innovations will be incorporated into supercomputer procurements that enable the European technologies to mature in real-world customer environments. Such an effort will also provide the opportunity to enrich the long-term European exascale technology roadmap, allowing HPC players to understand the overall vision of the EU HPC leadership program and where they fit in.
Finally, the development of the HPC ecosystem requires the joint effort of the supply and demand sides, with the aim of best matching the real needs of HPC users. The realisation of Europe as a world-leading innovation hub requires the new world-class computing capability to be easier to access and use in a wider European framework. This action is complementary to the actions points developed under "Governance" above.
Share with your friends: |
