Extreme computing: technological innovation and European leadership

EXPERT VOICE

Bull R&D at the heart of European projects and competitiveness clusters  

Technological innovation? It’s the very foundation of Bull’s existence, from the world’s first mainframes in the 1960s, to the invention of microprocessor boards in 1979, to globull™, the world’s most secure mobility platform launched in 2008 and now bullx, a new-generation of ultra-dense, ultra-modular supercomputers, designed with particular care.
All this goes hand in hand with Bull’s proven experience of worldwide collaborations with its technology partners (through its R&D centers in Europe, America and Asia), as well as with Open Source Communities, major industrial companies, prestigious research laboratories, and innovative small-to-medium sized enterprises as part of numerous European research programs and competitiveness clusters.     

Bull Direct interviewed three research directors, who talk about the high-level cooperative research projects they are involved in. Each illustrates just how high performance computing is making inroads in every sector of the economy, for businesses of different sizes.  

The HiPiP (High Performance Imaging Processing) project, with Jean-Claude Bourhis Launched in October 2008, this ITEA2 European Union project brings together partners involved in the French System@tic competitiveness cluster: Bull, the French Atomic Energy Authority (the CEA), IMSTAR and DOSISOFT, and the Dutch cluster Point-One: Philips Healthcare, Technolution and FEI, who have been chosen for their expertise in high-performance image processing in the field of bio-medical imagery.

Coordinated in France by Bull, the project draws on High-Performance Computing (HPC) technologies to support advances in medical image processing. The objective? To show the advantages these technologies offer when it comes to establishing a reliable diagnosis, applying radiotherapy or monitoring a treatment in real time, and even analyzing biopsy tissue samples at cellular level.  

The potential is huge, and there are numerous possible applications. What they all have in common is the need for increasingly precise 4D images (three spatial dimensions, plus time) for the different types of therapy used (X-rays, PETX, TEP¹, IRM², TDM³). Because the imagery is so much more precise, a great deal more data has to be processed in the short time that the patient is actually present during a consultation or examination. This is one reason why Bull has suggested standard multi-core systems packaged in different configurations (‘one box’, server, cluster) and helps partners to set up their applications in parallel so they can benefit to the maximum from the inherent power of their configuration.

Here are a few examples of possible applications, which illustrate the relevance of these technologies:

• Computer-aided surgical procedures. The aim is to eliminate the ‘noise’ present on the image in real time. This noise is caused by lower doses of the ionizing products to which patients are exposed; for example during a procedure to implant a stent (the weaker the signal is, the greater the interference and the more the noise increases).
• Computer-aided diagnosis, providing the practitioner with information just minutes after a clinical examination: information which would currently require hours of processing time (the reconstruction of 4D images from PET images, by analyzing a series of images taken sequentially, or the reconstruction of an image showing the cortical folds following an MRI scan). 
• Real-time radiotherapy simulation to assist practitioners with medical procedures. This simulation works as an interactive aid during radiotherapy treatment, with MRI, PET or TDM images being analyzed directly to help determine the ideal position and shape of the radiation beams. This means that non-target organs at risk of being irradiated can be avoided, that minimally invasive interventions can be visualized in real time, and very precise restitutions and rapid analyses of malignant cells during a sampling procedure can be carried out. 
• Early detection of disease with the construction of cognitive models to enable early detection of certain illnesses such as cancer (using multiplex cellular analysis) or neuro-degenerative illness’s such as Alzheimer’s disease.  

The project will deliver demonstrators for all types of treatment (X-rays, MRI, PET and CT-SCANs⁴ and cellular imagery) and an innovative stand-alone workstation for virtual simulation in radiotherapy.  
On a technical level, the project is particularly interesting in its use of application parallelization techniques and real-time applications. 

3D imaging of a brain's fiber networks with MRI

 

The EXPAMTION project with Jean-Marc Morel EXPAMTION is all about trialing a shared simulation infrastructure to serve everyone involved in the mechanical and electronics engineering (mechatronics) design cycle. Accredited by two French competitiveness clusters, MOV’EO and SYSTEM@TIC, the project was launched in December 2008 as part of MOV’EO’s DAS6 mechatronics project.

The DAS program is designed to boost the competitiveness of automotive parts manufacturers by implementing a collaborative design workshop bringing together professionals from the whole range of electronic and mechanical engineering disciplines.

The challenge for EXPAMTION is to speed up the development of innovative, environmentally-friendly products using high-performance computing resources that are shared in a totally secure way, ensuring that everyone involved has a practical way to work on developments cooperatively without any risk to their intellectual property. 

Alongside Bull – as the supplier of the highly secure cooperative HPC environment –EXPAMTION brings together parts manufacturers such as VALEO Systèmes Thermiques and MDP (a French SME), specialists in simulation applications such as ALTAIR Engineering, CADLM, INTES and SIMPOE, and the research institutes at CETIM⁷, the University of Versailles-St Quentin, and Compiègne Technology University. By bringing together all the required skills (mechanical and thermal scoping, electromagnetic/computational (CEM) compatibility, electronics, routing, etc.) across the whole design cycle, the workbench being developed and tested as part of this project will allow complex, ‘greener’ products to be developed and optimized. 

All those involved in the supply chain will test out new collaborative design methods, based on highly intensive use of multi-disciplinary simulation to develop products that, for example, can cut CO2 emissions by between 2% and 6% by optimizing weight-distribution of structures, heat-exchange and cooling systems, and electronically-controlled ventilation. 

This project will also address the issue of training, especially for SMEs in the sector, to ensure widespread uptake of the simulation tools, as well as defining a ‘business model’ for a secure collaborative simulation tool that can be shared between different industrial and academic players. The project also aims to promote the methodology and resources needed to develop this kind of cooperative venture in other sectors with similar supply chain issues (such as the aeronautical, mechanical and chemical industries).     
On a technical level, the project will also enable Bull to fine-tune its HPC solutions for mechatronics, optimize the modeling and simulation applications involved for its own infrastructures, and further enhance its security solutions, particularly the encryption speed for both stored data and data transmitted via networks.     

The TSAR (Tera-Scale ARchitecture) project with Huy-Nam Nguyen Accredited at the end of January 2008 by CATRENE – the European micro-electronics research program – TSAR is a project focused on designing the multi-core processor architectures of the future, aimed at the massively-parallel computing market.

 

Bull is the leader of this project, which also involves: the Thales group and the LIP6 laboratory of the Pierre & Marie Curie University on the French side; ACE, Compaan, Liacs, NXP Semiconductors, Philips Healthcare, Tu Delft on the Dutch side; and the FZI⁸ IT research center based at Karlsruhe in Germany.

TSAR is a project of critical importance for the years to come because today’s industry-standard chips are reaching their technological limits in terms of performance and energy consumption. Therefore, massively-parallel architectures are the only real way forward if we are to deliver the power needed by the latest computer simulation and HPC applications. 

The project – which involves pioneers in chip-based multi-processor networks (NXP Semiconductors, the Pierre & Marie Curie University/LIP6) – has three main aims:

• To design 1,000-core processor architectures capable of delivering around one Teraflops of compute power and improve the consistency of system caches using hardware-based mechanisms that are more powerful than the current software-based approaches. 
• In parallel, to develop applications to manage the parallelization and compilation of applications, as well as for porting operating systems to a shared memory multi-core platform.
• Lastly, to develop a virtual prototype (at the transactional level) to run a number of benchmarks and applications, and evaluating their performance in order to validate that this type of architecture is appropriate. 

On a technical level, effectively managing these card-based multi-processor architectures (MPSoC⁹ ) and integrating them into HPC solutions is vitally important for Bull, with its ambition to be market leader in computer simulation in Europe.  

HPC – Bull’s involvement in European research programs and competitiveness clusters