Section: Other Grants and Activities
ADT Galaxy (INRIA transversal action)
The ADT Galaxy contributes to make INRIA a value-added player in the SOA arena, by designing and developing an Open Framework for Agile and Dynamic Software Architecture. This ADT will work for INRIA and INRIA's research project-teams direct benefit, and aims at pre-assembling technological bricks from various teams, projects and preparing them to be transferred through the open source software channel.
The ADT aims at providing an IT agile platform, built on dynamic software architecture principles, and fitting for flexibility, dynamical reconfiguration, adaptability, continuity, and autonomic computing. Fractal, SCA-Tinfi and GCM/ProActive are the major technological drivers of this ADT. The different usage scenarios as well as the different tools developed at infrastructure, application, and business levels demonstrate that this platform is able to support the design, modelling, deployment, and execution of business processes. In the same time, the ADT targets the definition of a new common language to manipulate dynamically adaptive distributed SOA-based systems, encompassing application, and middleware layers. This common language will take different forms, inherited from works done by several project-teams with their distinct skills, and illustrates a new kind of collaboration between teams, coupling research and development works.
Contributors to this ADT are mainly research project-teams, including OASIS, ADAM (Lille), ECOO (Nancy), ASCOLA (Rennes), ObjectWeb/TUVALU (Grenoble), SARDES (Grenoble) and TRISKELL (Rennes), and the ADT Galaxy is led and managed by the TUVALU team.
The duration of this ADT is over 28 months : the kickoff meeting has been held on July 3rd, 2008 and the project is planned by end of October, 2010.
This ANR funded project gathers partners that are applied mathematicians (OMEGA/NACHOS and SMASH teams), and computer scientists researching in distributed and grid programming environments (OASIS, PARIS, LaBRI SoD, MOASIS).
The DiscoGrid project aims at defining a new SPMD programming model, suited to High Performance Computing on Computational Grids. Grids are hierarchical in nature (multi-CPU machines, interconnected within clusters, themselves interconnected as grids), so the incurred latency for inter processes communication can vary greatly depending on the effective location of the processes. The challenge is to define a programming model that allows programmers to exploit this hierarchy, as easily and efficiently as possible. As the MPI SPMD message-passing model is very popular in High Performance Computing we are defining a hierarchical extension of MPI.
The DiscoGrid project is developed upon four main axes: the definition and implementation of a grid-aware mesh partitioner; the specification of a high-level MPI-based programming interface; the design and development of a runtime supporting the proposed extensions and inter-cluster communication and the development of real-size simulation software based on partial differential equations (PDEs).
Initially, the Oasis group was involved in the specification of the programming interface with the definition of the main concepts of hierarchical SPMD communication. In a second phase, the group developed the programming interface (C/C++ and Fortran) and the infrastructure needed for native code wrapping as well as a prototype based on Active Objects and adaptation of existing numerical applications, namely the Poisson3D solver and BHE  ,  ,  .
This project brought important contributions to the GCM, these contributions include: better definition of collective interfaces; optimisation techniques for collective interfaces; and wrapping and deployment of native code applications, including MPI. Several publications showed the importance of a grid-oriented approach on the development of non-embarrassingly parallel applications.
This project started in January 2006, initially for 36 months, involving 110 kEuros. The project was extended for 6 months upon a demand of project partners, ending in June 2009.
The aim of this project is to define, develop and experiment a collaborative platform of mutidisciplinary optimization - As "platform" we consider here a software environment hosting heterogeneous code and data, geographically distributed in equally heterogenous machines. These codes can be sequential or parallel. These machines can be data servers, supercomputers, PC farms, etc. - As "collaborative" we consider this environment to be able to host, control and allow communication of these codes transparently for the user, according to their own work habits. - As "multidisciplinary optimization", we consider the collection of methods and numerical tools, objects of the other tasks in OMD2 and previously realized during project OMD. It has been decided that ProActive will be used as middleware for communication between machines, and that the ProActive Scheduler will be used as a basis for the collaborative platform. Similarly, Scilab will be used as a common language to describe optimization strategies.
The project started January 1st 2009 and will last 3 years, with a total budget of 214 kEuros.
Contrat Plan État Région Grille et Calcul Pair-à-Pair
This contract aims at building a regional computing platform. This is achieved by mixing desktop machines with dedicated ones like clusters. Users willing to submit a job will do so by accessing a web-page and uploading their program. It will then be scheduled and executed on a free machine. The scheduler is currently under development.
In the first part of the project, the access to the platform will be restricted to Inria members. Once most of the tools have been developed, the access will be open to industrial partners.
A convention has been signed with Microsoft to provide a specific cluster with Microsoft Compute Cluster Server.
The members of this project are the Inria and the Eurecom institute (Télécom Paris Ecole Polytechnique Fédérale de Lausanne).
The total budget for this project is 500kEuros for Inria and 100kEuros for Eurecom.