Quest for the real-time for the safety analysis code Cathare 2 used in the post-accident simulator Sipa

The aim of the SCAR project is to use the CATHARE French thermal-hydraulic accident code in the SIPA simulator (Post-Accident Simulator) and extend SIPA to reactor cold shutdown states. The quest for real-time has been one of the key themes of the project since it began in 1997. The required CPU time depends on the computing power and on the ability of CATHARE to converge as fast as possible on the solution. Three main tasks have been scheduled to contain the lag between the simulation and the real-time: -1) Parallelism in CATHARE has been developed with shared-memory model (using OPEN MP). Standardized and adapted to the numerical method and the structure of CATHARE, it has enabled parallel tasks in 95% of the code with efficient parallel loops on the elements, and an optimized but limited parallelism in the solver. Validation has been carried out all along the task, ensuring the binary identity of results for 10 representative accident transients, whatever the number of processors used on each computer of the SCAR project. -2) Convergence has been improved for 20 CATHARE transients, ranging from the 100% full power state to cold-shutdown for maintenance state. A method based on the definition of maximum lag criteria in function of an estimated power of computers has been developed, revealing coding errors and leading to numerical improvements without any regression of physical law validation. A second phase has started in 2003 on another series of 25 transients within the simulator. -3) A techno-watch policy (using benchmarking) has allowed to keep up to date with progress in computer power throughout the duration of the project. It has consisted in comparing the performance of computers for 12 standard CATHARE input decks using an elementary time relevant of the computing machines for a given modeling of plant series. Furthermore, development validation and performance assessment tools have been developed at the same time. As a result of these three tasks, CATHARE is now run in parallel inside the simulator, modeling the primary, secondary, and Residual Heat Removal Systems, the presence of three non-condensable gases, as well as boron, iodine-131, and nitrogen-16 transport. With the present computers, real time is effective for operating and small breaks transients, but larger breaks and cold-shutdown state transient still remain handicapped by too many iterations. The progress in computer power would indicate that the hypothesis of an elementary time of 10 μs, adopted in 1997, would soon be confirmed. (authors)