As key strategy levers, High-Performance Computing (HPC) and simulation are helping researchers to overcome major scientific and technical hurdles, and industry to become more competitive through being able to design, develop and optimize new products more rapidly.
In future years, we will be seeing what we now consider typical power levels concentrated into systems with ten times the processing capacity.
Read on to find out how Bull, with the FAME2 customer-driven project and partners from the System@tic competitiveness cluster, is building the HPC solutions of the future. You may also discover these solutions have a great deal to offer your organization.
What new avenues are being opened up by research into the field of high performance computing?
High Performance Computing (HPC) has seen prodigious growth over the past few years: The most powerful supercomputer installed to date is equivalent to the combined power of the 500 largest systems installed just six years ago. Tomorrow’s challenge is to take this one stage further, since while the power available today offers substantial advances to industry, as well as to the research sector, some simulations are still not achievable. And this explains the driving force behind the race to achieve maximum computing power. Currently, the most powerful supercomputers in the world have a capacity of several hundred teraflops1 . Tomorrow, the objective will be petaflop2 capacity , effectively ten times as powerful! Such an increase will significantly speed up industrial design, applied research into petrochemicals, integrative biology… but will also play its part in handling the huge volumes of multimedia data that will be generated by Internet applications open to millions of Web users.
Bull is investing heavily in this area, in close collaboration with customers, industrial partners and research laboratories at the System@tic Paris-Région3 competitiveness cluster, in the framework of the FAME24 innovation program.
What is the aim of the FAME2 program?
The program aims to ensure that our supercomputers – which will be petaflop capacity machines – are built to fulfill the future requirements of HPC as well as multimedia databases, and that they are delivered along with a comprehensive portfolio of applications, libraries and optimized tools. As ‘Architect of an Open World™’, we have launched the FAME2 project with our partners so as to:
- Anticipate software and application development requirements
- Extend the range of application areas covered
- Ensure these supercomputers will be able to manage very large databases optimally.
One of the project goals is to find out how to exploit the very high degree of parallelism represented by more than 100,000 computing cores, most efficiently.
In the framework of the System@tic cluster, our partners are validating these supercomputers with their most advanced applications, and are anticipating extensions or upgrades to them to enable them to extract maximum power from their architecture. The annual workshop, which takes place in mid-October, has enabled us to evaluate the overall development and testing program with them, and I am pleased to say that FAME2 is progressing absolutely to plan.
What, in your view, is the main advantage offered by the competitiveness cluster?
System@tic has enabled us to considerably increase the level of collaborative work around FAME2, which was initially launched under the auspices of Ter@tec5, an initiative driven by the French Atomic Energy Authority (the CEA). This has led to us bringing together players from the world of industry such as the CEA itself, Dassault Aviation, the French Petroleum Institute (the Institut Français du Pétrole), ILOG, young innovative companies such as NewPhenix (multi-criterion research), CAPS Enterprise (parallelization tools) and Resonate MP4 (multimedia), and research laboratories including ECP/MAS, INRIA/IRISA, INT/Artémis, IBISC, UVSQ/ITACA. The result is a powerful ecosystem enjoying the support of regional and national authorities, and an excellent dynamic for co-operation and innovation. The appeal of the cluster is such that co-operative ventures with other players interested in the project are taking shape, and these extend well beyond the formal contractual framework of FAME2.
Bull is also participating in three other projects in the System@tic cluster:
- CARRIOCAS6 or CAlcul Réparti sur Réseau Internet Optique à CApacité Surmultipliée, a project looking to meet the computing and display requirements for interactive digital or virtual reality distributed applications (virtual factory)
- SIC or Sécurité des Infrastructures Critiques, a project aiming to resolve the security issues of critical infrastructures such as airports or railway stations, that have highly concentrated resources and vast numbers of people on the move
- PFC or Plates-Formes de Confiance, a project aiming to develop the primary technologies for trust and sovereignty in information systems security.
In addition to the above, we have committed to several other projects due to be launched during 2007, the most notable being personalized TV channels, featuring dynamic enrichment of video content and feeds.
Can you tell us a bit more about the applications currently being optimized?
Before we look at applications, I would like to talk a bit about our work on very large databases, because all high-performance computing applications manipulate and generate huge volumes of data (up to hundreds of Terabytes with a very high number of access events). To cite just one example, the CEA’s TERA-10 supercomputer generates up to 10TB everyday. Therefore, we have identified the very large databases in XML format and multimedia flows, on which we are working with the CEA, Résonate MP4, INT/Artémis and NewPhenix, as particularly representative of emerging applications.
Returning to applications, the emergence of supercomputers has opened the way to a level of digital simulation bringing us closer to being able to faithfully reproduce very complex phenomena. These new applications are strategically crucial to resolving today’s key challenges; they will have a huge impact on how we live in the decades to come.
Before handing over to our partners, I would just like to briefly describe some of them:
- New product design, and notably Dassault Aviation’s aerodynamic simulation of air flows around its Falcon business aircraft
- Researching new energy sources by simulating, for example, oil reserve exploitation by the IFP using models comprising several tens of millions of grid cells
- Numerous applications in life sciences. To mention just a few, I will take the example of integrative biology that requires both integrated and multiple-scaled modeling of biological processes and integration of biological data which is as varied as it is abundant; two fundamental issues being taken on by the IBISC laboratory at the University of Evry. Among other examples of modeling applications are the study of how cancer cells behave, processing simulations, simulation of interactions between a virus and its ‘host’, molecule modeling used in the design of new medicines, etc…
- System failure simulation. In this domain, we have recently concluded a partnership with the ESI Group for its PAM-CRASH application, designed to strengthen security right at the design stage for new automobile models
- Climatology and seismic prediction of natural catastrophes
- Staff and vehicle scheduling for major transport companies
- Multimedia information research, natural language, business intelligence etc.
What, to sum up, are the main distinguishing features of these future petaflop systems?
Openness, balance, and power.
Openness, because they are based on standard processors, supplied in volume by Intel. Their entire Open Source-based software environment is similarly optimized under Linux in the framework of the FAME2 project.
Another characteristic is that computing power and input/output power are balanced, and this makes them very powerful systems both for HPC and for large databases, as well as for the most demanding of business applications.
FAME2 : innovation et simulations numériques. Commentaires de partenaires
1 Teraflops : 1012, one thousand billion operations per second; teraflop derives from the Greek word τέρας, denoting monster
2 Petaflops : 1015, one million billion operations per second; petaflop derives from the Greek word πέντε, five because it is equal to 10005
3 At the heart of the digital revolution, the global competitiveness cluster SYSTEM@TIC PARIS-REGION brings together nearly 200 industrial players, academics, and French institutions around collaborative R&D projects spread over four markets with high added-value application: Telecoms, Security-Defense, Automotive-Transport, Systems Design and Development Tools. Collaborative research/industry projects launched from the SYSTEM@TIC PARIS-REGION cluster typically embody this kind of innovation dynamic, which is a key decisive factor when it comes to industrial competitiveness.
4 FAME : Flexible Architecture for Multiple Environments, the architecture designed by Bull, and used in its NovaScale® servers.
5 Created by the CEA and backed by regional authorities, Ter@Tec is a structure for exchanges and collaboration between the various players involved in high-performance digital simulation: researchers, computing and industrial companies.
6 The aim of the CARRIOCAS project is to study and develop an ultra-high flow fiber optics network (40 Gigabits/s per channel) with access to very high capacity (10 Gigabits/s and more), capable of responding to industrial needs in terms of interactive digital simulation on remote supercomputers and handling very large volumes of remote data.