OASIS4 validation¶

The situation has evolved rapidly in the last weeks. Since January, we were trying to freeze the first official release of OASIS4. We did a lot of work introducing multi-block partitions and revisiting completely the interpolation global search for the Gaussian reduced grid. We did a lot of testing and unfortunately, each new case revealed new problems. Even if this is not satisfactory, we finally decided last week to merge the development_4 branch on the trunk and call this oasis4_1beta. All remaining problems we know about are listed into TRAC tickets (see https://oasistrac.cerfacs.fr/report/1). What we were suspecting since a while became really clear in the past weeks: the OASIS4 parallel interpolation global search is not a good basis to build on; it got to a point where it is too complex to be evolutive.

The plan for the short term is to test, validate and hopefully use the OASIS4 user-defined interpolation which completely bypasses the parallel global search, for "high-resolution" coupled models. The interpolation weights and addresses are pre-calculated off line by the user, they are read in by OASIS4 which performs the matrix multiplication and the parallel redistribution of coupling data between the source and the target. This is the first fall-back solution to have a parallel coupling tool for our high-resolution runs, removing the current OASIS3 bottle neck. The second fall-back solution that we also started to investigate is to interface the american MCT (Model Coupling Toolkit, www.mcs.anl.gov/mct) directly in OASIS3; this should be technically equivalent to the "OASIS4 user-defined interpolation" in the sense that MCT also read pre-defined weights, performs the matrix multiplication and the parallel redistribution of coupling data between the source and the target.

For the off-line calculation of the weights and addresses, we are currently evaluating some new ESMF functions. They are supposed to give better and faster results than the SCRIP library currently used in OASIS3 and OASIS4, which shows specific problems near the poles.