The meeting took place in CERFACS in Toulouse. The WP3a participants were: from NEC-CCRL, Rene Redler and Hubert Ritzdorf; from SGI Deutschland, Reiner Vogelsang; from FECIT/Fujitsu, Jean Latour and Arnaud Caubel; from CERFACS, Damien Declat and Sophie Valcke; from MPI/M&D, Stephanie Legutke and Hannes Thiemann. The WP4a participants were: from IPSL, Marie-Alice Foujols, Philippe Bourcier and Alexandre Tallec; from the MetOffice: Steve Mullerworth and Mick Carter . Other participants were: from ECMWF, Claes Larsson; from KNMI, Peter van Velthoven; from MPI, Reinhard Budich (PRISM Director); and from the University of Manchester,  Rupert Ford and Graham Riley who are consultant in the MetOffice FLUME project.

The objectives of the meeting were to discuss the coupled system general architecture, to detail the common coupling and I/O model interface library, to identify the different development tasks and to define the responsibilities of each partner.

General overview of the coupled system

A general overview of the coupled system was first presented by Alexandre Tallec and discussed. The proposed architecture supposes a common coupling and I/O model interface library, the "PRISM System Model Interface Library" (PSMILe). The participants made the following remarks:

• Conceptually, no difference should be made between a model and a file: both are entities that may provide data to another model. The difference arises at the deployment phase in which different implementations of the "entities providing data" may be chosen (file or model). One can argue here that there is a conceptual difference in the sense that a file is an entity containing data unable to perform a providing action (the data will have to be fetched), while a model is an entity containing data able to provide them to the "external world".
• If no conceptual differences is made between a file and a model, there is no reason at composition phase to differentiate between the SMIOC and the SCC. All composition information should be gathered into only one container. For practical reasons, the information relevant for each model could then be re-dispatched to each model.
• The content of the container giving for each model a description of its interfaces, the PMIOD, was also discussed in detail. The PMIOD should contain metadata giving a description of the related model, its possible grids, and its input and output data. The concerned input and output data were finally divided into 3 categories. These 3 categories should be described separately in different sections of the PMIOD or in separate containers, and should be treated differently by the PSMILe.
• Transient input and output variables: data evolving during the run, that will be received or provided by the model at an a priori unknown frequency, from or to an external entity (the frequency will be defined by the user during the composition phase). For output data, the external entity may be the user requesting diagnostics.
• Restart input variables: data requested initially by the model to (re)start the simulation and provided previously by the model itself (not by an external entity).
• Model input parameters: data requested initially by the model to define the physical configuration of the experiment. These parameter may have default values but may be adapted by the user. For coupled models, the "global model input parameters" are parameters that may require a consistent definition among all the model components.
• One same model code may implement different interfaces at deployment phase. It should be possible to describe these different interfaces and the conditions under which they are realized in the PMIOD.
• At composition phase, the user assembling the system (and therefore defining the SMIOC and the SCC) will need some deployment information, not present in the metadata described in the PMIOD. For example, the user may need to know the resolution at which the model will run in order to choose the appropriate interpolation.

Alexandre Tallec then presented a first proposal for the metadata associated to the model, its possible grids, and its input and output data (link to the presentation ???). The following issues were discussed:

• A definition of the metadata through a DTD structure written in XML was proposed for the PMIOD, SMIOC and SCC. Philippe Bourcier and Alexandre Tallec will verify if a F90 interface to XML is available.
• In general, the word "vector" should be used instead of "flux".
• For transient input/ouput variables, the "origin/destination" should not be included in the metadata, as the model usually ignores it. More work is required to evaluate if the minimal/maximal frequency should be included. If included, it should probably be expressed in terms independent of the deployment characteristics (for example, in terms of timesteps but not in terms of seconds).

Presentation of IOIPSL current interfaces and I/O related issues

Philippe Bourcier and Marie-Alice Foujols then presented an example of IOIPSL current interfaces based on the HISTCOM package. IOIPSL is the I/O library currently developed at IPSL.

The issue of parallel I/O was discussed. It was mentionned that MPI-I/O could be adapted to the netCDF format; Hubert Ritzdorf and Philippe d'Anfray could be advisors on this point. It was also mentionned that Balaji's work at GFDL on parallel I/O should be considered.

It was also specified that the I/O library will produce and consume netCDF files. Clear interfaces to produce or consume other formats will be provided, but the routines producing or providing other formats will not be coded by WP4a.

For diagnostics, the answers to REDOC questionnaire I.5 will have to be analysed. This questionnaire should be distributed to the PRISM partners in the coming weeks.

Definition of tasks and repartition among the partners

The different PSMILe functionalities, actions and external interactions are summarized in the following table:

The tasks associated to the development of the different parts of the PRISM coupled system were defined and distributed among the partners as follows:
 
 

Task	Leader	Date
T1 -  Detailed content of PMIOD, SMIOC and SCC	Alexandre, Sophie	01/05/02
T2 -  Metadata management (localisation, use, transfer, ..)	Alexandre, Sophie	01/05/02
T3 -  Metadata DTD definition + two examples	Alexandre	01/09/02
T4 -  Definition of PSMILe interfaces (D2c3)	Rene, Marie-Alice	01/06/02
T5 -  Detailed list of PSMILe internal elements	Sophie, Marie-Alice	01/06/02
T6 -  PSMILe software data structure definition	Reiner, Philippe	01/07/02
T7 -  Flow chart for PSMILe F1, F2, F3, and F6 routines	Hubert, Rene	01/07/02
T8 -  Flow chart for PSMILe F4, F5, F7 routines	Marie-Alice	01/07/02
T9 -  Flow chart for PRISM driver	Damien, Sophie	01/07/02
T10 - Flow chart for non-local transformation entity	Damien, Sophie	01/07/02
T11 - PSMILe internal elements interface definition	pair of element developers	01/09/02
T12 - PSMILe for OASIS 3.0 (D3a1)	Sophie	01/10/02
T13 - PSMILe for I/O library prototype (D4a1)	Marie-Alice	01/10/02
T14 - Merging of PSMILe for OASIS 3.0 and for I/O library prototype	Sophie, Marie-Alice	01/11/02
T15a - Provision of an PSMILe test program	Marie-Alice, Sophie	01/11/02
T15b - Provision of an PSMILe benchmark based on test	Reiner	01/12/02
T16 - F90 XML access	Philippe	01/01/03
T17 - Programming of PSMILe F1, F2, F3 routine(s)            - PRISM coupler prototype(D3a2)            - PRISM coupler final version (D3a3)	Rene, Hubert, Philippe  Rene, Hubert, Philippe	01/09/03 01/09/04
T18 - Programming of PSMILe F6 routine(s) including DEL            - PRISM coupler prototype(D3a2)            - PRISM coupler final version (D3a3)	Rene, Hubert,  Rene, Hubert	01/09/03 01/09/04
T19 - Programming of PSMILe F5, F7 routine(s) (D4a3)	Marie-Alice	01/09/03
T20 - Programming of PSMILe F4 routine(s) (D4a3)	Philippe	01/09/03
T21 - Programming of PSMILe I/O routines (D4a3) for F5, F6, and F7	Philippe	01/09/03
T22 - Programming of coherence check routines -> prototype                                                                                  -> final version	Jan  Jan	01/09/03 01/09/04
T23 - Programming of local transformation routines -> prototype                                                                                         -> final version	Sophie Sophie	01/09/03 01/09/04
T24 - Programming of PRISM coupler driver            - PRISM coupler prototype(D3a2)            - PRISM coupler final version (D3a3)	Damien Damien	01/09/03 01/09/04
T25 - Programming of PRISM coupler non-local transformation entity            - PRISM coupler prototype(D3a2)            - PRISM coupler final version -parallelisation (D3a3)	Damien Damien, Thomas	01/09/03 01/09/04