Section: Application Domains

Outline

The scientific methodology of MOAIS consists in:

• designing algorithms with provable performance on generic theoretical models. In particular we develop randomized algorithms for distributed scheduling and approximate multi-objective optimization theory..

• implementing and evaluating those algorithms with our main softwares:

  • Kaapi for fine grain scheduling of compute-intensive applications;

  • FlowVR for coarse-grain scheduling of interactive applications;

  • TakTuk, a tool for large scale remote executions deployment.

  • Triva, for the visualization of heterogeneous parallel executions.

  • KRASH, to generate reproducible CPU load on many-cores machines.

• customizing our softwares for their use in real applications studied and developed by other partners. Applications are essential to the validation and further development of MOAIS results. Application fields are: virtual reality and scientific computing (simulation, visualization, combinatorial optimization, biology, computer algebra). Depending on the application the target architecture ranges from MPSoCs (multi-processor system on chips), multicore and GPU units to clusters and heterogeneous grids. In all cases, the performance is related to the efficient use of the available, often heterogeneous, parallel resources.

MOAIS research is not only oriented towards theory but also focuses on applicative software and hardware platforms developed with external partners. Significant efforts are made to build, manage and maintain these platforms. We are involved with other teams in four main platforms:

• SOFA, a real-time physics simulation engine (http://www.sofa-framework.org/ ;

• Grimage, a 3D modeling and high performance 3D rendering platform (http://www.inrialpes.fr/grimage ) and its evolution with the new Kinovis platform.

• Digitalis, a 780 core cluster based on Intel Nehalem processors and Infiniband network. Digitalis is used both for batch computations and interactive applications;

• Grid'5000, the exprimental national grid (http://www.grid5000.fr/ ).