RAPID

NETWORK OF RESOURCES CENTRES FOR INDUSTRIAL PARALLEL APPLICATIONS AND SCIENTIFIC DEVELOPMENT

A joint initiative of CEA - CERFACS - CNRS - DRET - INRIA - MESR

Last update: april 19th, 1996.

CEA :: Commissariat à l'Énergie Atomique
CERFACS :: Centre Européen de la Recherche et de la Formation Avancée en Calcul Scientifique
CNRS :: Centre National de la Recherche Scientifique
DRET :: Direction des Recherches, Études et Techniques (Ministry of Defense)
INRIA :: Institut National de Recherche en Informatique et Automatique
MESR :: Ministère de l'Enseignement Supérieur et de la Recherche

The main objective of the RAPID consortium is to promote and transfer the know-how of high performance parallel computing to the industrial and academic community users or future users of intensive computing.

Motivations

Parallel processing has clearly become a key economic and strategic issue. This is illustrated by the fact that parallel computers have reached the business market and by the increasing number of applications in need of this new technology. All prospective studies agree that- by the year 2000, high performance computers will be based on parallel architectures. Currently, parallel machine technology is undergoing rapid evolution phase, yet the simple, user-friendly interface needed to generalise their use has not been designed yet. One of today's difficulties is the lack of standards and tools. The market expects that semi-automatic methods of parallelization will enhance the capacity of these machines. This is why, many potential industrial and academic users are reluctant to invest in this technology.

RAPID brings together French research centres specializing in parallel computing. Its aim is to disseminate and transfer the know-how in this field. Our informatics and applied mathematics laboratories have been working for more than 20 years in the areas of distributed and intensive computations, and for more than 10 years on high performance parallel architectures.

The RAPID network includes computing facilities which are located in Bordeaux (LABRI), Grenoble (CEA and IMAG), Lille (LIFL), Lyon (LIP, PSMN and CDSCP), Marseille (IRPHE), Montpellier (CNUSC), Nancy (Centre Charles Hermite ), Paris (CNCPST, IDRIS and ETCA), Rennes (INRIA/IRISA), Rocquencourt (INRIA), Rouen (Crihan), Saclay (CEA), Sophia (INRIA), and Toulouse (CERFACS). The computers are accessible via the high performance electronic network RENATER. An additional list of centres will be published at a future time.

RAPID uses a simple and efficient structure to coordinate parallelization activities in France with the aim of promoting a real synergy between research and industry.

Available Services

RAPID offers potential users :

hardware and software infrastructure for parallel computing, accessible through public high-speed communication networks,
expertise on the choice of an architecture or a programming model, a programming language or a numerical method,
technical support by a team of engineers specialised in porting, optimisation or application development, in the wide range of parallel machines available such as : SGI Power Challenge Array, Kendal Square Research, Intel Paragon, TMC-CM200, CM5, MEIKO CS-2, IBM-SP2, CRAY-T3D, CONVEX-SPP, etc...
a variety of training courses covering all areas of interest in the field, ranging from parallel programming to the latest technological sofware or hardware developments.

The parallel machines available in RAPID's research centres are coupled with technical assistance to overcome first-approach problems, like code analysis and parallelization concept definition. Expertise on a large number of fields also allows RAPID to offer services varying from training courses to the porting of applications or a new project development. Active collaboration between users and founders of the network is encouraged.

Principles of action

Two kinds of actions which can be undertaken, correspond to two successive phases :

Feasibility studies : to evaluate the saving in execution time and the work load that the porting of a large code onto a given parallel machine would represent, a company can access one of the nodes without being charged for the initialisation of the work. Access to the machine and help from services in the centre is granted up to a certain level of occupation time.
Porting of applications: some feasibility studies specify the need for application portability. In this case a contract is drawned up between the owner of the code and a network node that sets up a task force group in charge of the work.

Composition of the Network

The structure and status of RAPID are defined an agreement between its partners. This agreement specifies that the network is supervised by a Scientific Committee composed of representatives of each node plus externally invited experts. The network management is conducted by a Management Committee following the same principle.

The role of this committee is to centralize demands for collaboration and to give a quick response, indicating the most appropriate centre and organizing the work to be done.

The offer is not limited to the centres of the contractors, but it will also involve associated centres which are known to have valuable experience in parallel computing.

Special emphasis is placed on conducting adequate information, marketing and training towards potential users, in particular industrial companies.

Contact