I. Summary

The content of this project is twofold: Make use of high speed networks for distributed computing in term of network optimization, and realize efficient coupling between different CFD codes for the treatment of complex flows phenomena. This proposal is foreseen as an application project of the Internet2 initiative in its international extension with France.

II. Project Description

1) Metrology and optimization

The CFD laboratory of the Indiana University has developed efficient tools to evaluate the load balancing requirements for optimizing block-structured CFD solvers. That tool allows deciding what is the best domain decomposition strategy to port a multiprocessor solver onto a heterogeneous cluster platform. From a "probe" simulation, it is then possible to map in the most efficient way the software architecture onto a given hardware platform.

CORIA has developed a wide variety of computer codes in the field of combustion, multiphase flows and complex aerodynamics. Among the problems for which a good expertise is recognized, the treatment of coupled problems on parallel computer platforms has been extensively worked out during the last three years. Application of coupled problems were done for turbulent combustion in supersonic flows, complex compressible aerodynamics and fluid-structure interactions.

Nevertheless, implementation of these coupled solvers has been governed essentially in terms of physical models or raw computer efficiency, without paying too much attention to quantitative aspects of load balancing on heterogeneous hardware platforms.

As a first step for this cooperative project between Indiana University and INSA of Rouen, the load balancing evaluation tools from Indiana will be applied to two parallel solvers in use at INSA. The first one will be the coupled fluid-structure interaction code (application to the dynamic deformation of a rocket engine nozzle). The second one will be a intensive computing Navier-Stokes solver for complex aerodynamics problem (Application to airbreathing engine air intake aerodynamics).

2) Coupling

After that preliminary phase will be achieved, it will be possible to consider a distributed execution on hardware platforms distributed in both laboratories, linked with a high speed link provided on the US side by the Internet2 project (Abilene/UCAID network) and on the French side by the RENATER2 infrastructure.

Specific action will have to be done to calibrate carefully the network resources between the two platforms. On the US side, Indiana University will provide two cluster platforms interconnected with a high speed local network, and implementing either MPI or PVM as communication library above IP protocol. On the French side, a SGI Origin2000 computer with 64 processors and 32 gigabytes memory will be available. SGI platform will run PVM, MPI or Open-MP as internal communication library.

The proposed application for that second phase will be, either to distribute the execution or a very large scale configuration or to run a coupled aerodynamic problem. A typical configuration could be the air flow calculation along an air intake followed by the first rows of turbomachinery for a civil aircraft engine. The air intake could be treated with the aerodynamic code of CORIA, whereas the rotating component part will be treated with a specific code of the CFD laboratory of Indiana University.

III. Organization

Duration

The expected duration of that project is about 2 years. The first 6 months will be devoted to the metrology and load balancing evaluation. The following 18 months will be needed for the realisation of the coupled problems and their exploitation.

Type of exchanges

The basic exchanges will be for short term reciprocal hosting of researchers participating to the projects, and for the sustained availability of a high quality network link between the two laboratories.

Depending on the issue of the first phase, other collaborative platforms will be sought with the University of Stuttgart computing center (Dr. Michael Resch and Dr. Alfred Geiger) in Germany and with the University of Kyoto (Prof. Satofuka) in Japan.

Resources

Requested resources for this project are mostly to support the travel expenses, the project management and the network costs involved.