June 11 this year saw the creation of the joint laboratory between INRIA and the National Center for Supercomputing Applications (NCSA) at the University of Illinois in the US. This provides a unique opportunity for French researchers to team up with the Blue Waters project, which in 2011 will be the academic world's most powerful computer.
Interview with Franck Cappello, co-director of the Joint Laboratory for Petascale Computing.
: What does the creation of this joint laboratory represent for INRIA?
Franck Cappello: It's the first time that the institute has had the chance to join a major US project financed by the NSF1. The aim is to design software for what will be the most powerful computer in 2011, which means working with the best US teams and being at the cutting edge of work on high-performance computing. For their part, the US researchers are interested in our solid fundamental bases in parallel programming, digital libraries and fault tolerance, which are the joint laboratory's three research topics. Our association with the Blue Waters project is therefore also an acknowledgement of the skills developed at INRIA over many years in the field of high-performance computing, both in research and transfer.
: What scientific challenges does the laboratory face?
Franck Cappello: The laboratory's aim is to develop software tools enabling this supercomputer to operate at an optimal level. This means successfully achieving a sustained performance of one petaflop (a million billion calculations per second) in real-world applications and not just in tests. The challenge is especially interesting because Blue Waters is not just larger than previous machines, with its 200,000 processor cores, but also constitutes ground-breaking technology that is particularly stimulating for researchers. Designed by IBM, this architecture was chosen by DARPA2 at the end of a programme aimed at identifying a petaflop technology. What makes it unique is that it establishes a new balance between processor performance and memory and storage access capacities, these capacities having been increased.
Furthermore, faults will be a recurring and normal event with machines of this size. It is therefore essential to factor this phenomenon into our approaches so that we can ensure correct functioning. We have already published a technical report on fault tolerance, which is currently one of the two references on the subject at the site of the Los Alamos National Laboratory.
: Will European researchers benefit from the results achieved by this laboratory?
Franck Cappello: Being associated with this project means being one step ahead and working on machines that will arrive in Europe a few years later. This is important, because a good understanding of computer architectures is needed in order to develop outstanding software. This lead is particularly interesting because machines smaller than Blue Waters - and therefore more affordable - but with the same configuration are already being prepared at IBM. In addition, the software created in the joint laboratory will be available in open source format and can therefore be reused by the European scientific community for its own needs.
: What are your ambitions for the joint laboratory and for INRIA?
Franck Cappello: As co-director of the laboratory, I'm trying to establish a research climate that is favourable to the achievement of important results, that is, results that serve both the project and the reputation of the institute. I'm looking to establish the most effective synergies possible, notably by organising workshops that can identify teams interested in the topics addressed. The next workshop, in which around ten teams from the institute will take part, is already scheduled for December. Meanwhile, the US research environment itself creates the conditions for a highly stimulating collaboration, as researchers, operators at the computing centre and users are closely connected and form an integral part of the project. One of my goals is for at least 20 or so researchers from INRIA teams to come to the laboratory every year, whether for visits or longer assignments.
My long-term ambition is for INRIA to become a preferred interface for our US partners in high-performance computing, which for example would see it included in explorations into the next generation of exascale computers, scheduled for 2018.
(1) NSF: National Science Foundation (2) DARPA: Defense Advanced Research Projects Agency