1) User Survey results and follow-on (S. Valcke)
2) OASIS4 source code
reorganisation and compiling
environment (
3) SCE adaptation
4)
FCM
· The experience gained with the MetOffice compiling environment with Oasis3 during the September workshop was presented by Sophie (see FCM.pdf, ext_cfg.html.)
· It was decided that the FCM Perl scripts could be provided (given authorization by the Met Office) in OASIS4/util directory. But full pre-built Makefiles with adaptable headers, as in current version, will also be provided. The current compiling system philosophy is that the user does not recompile the code often so it is better to release a safe basic Makefile than a more sophisticated but potentially hard-to-debug Makefile.
5)
OASIS4
Users and applications:
· It is useful to keep track of users and coupled applications based on OASIS4. However no easy way to do this was proposed. A reminder to feed this information back to us could be sent every year to OASIS4 mailing list.
6)
Interpolation
and repartitioning issues:
1. Interpolations currently supported
2. Grids currently supported :
2.1 Regridding, repartitioning, I/O:
· Regular in lon, lat, vert (“Reglonlatvrt”): lon(i), lat(j), height(k)
· Irregular in lon and lat, regular in the vert (“irrlonlat_regvrt”): lon(i,j), lat(i,j), height(k)
·
Irregular in lon,
lat, and vert
(“irrlonlatvrt”) (not fully tested): lon(i,j,k), lat(i,j,k),
height(i,j,k)
· Gaussian Reduced in lon and lat, regular in the vert (“Gaussreduced_regvrt”): lon(nbr_pt_hor), lat(nbr_pt_hor), height(k)
2.2 Repartitioning and I/O only:
· “Gridless” fields: no geographical information attached; local partitions described in the global index space (prism_def_partition).
2.3 I/O only:
· Unstructured grids (“unstructlonlatvrt”): lon(npt_tot), lat(npt_tot), height(npt_tot)
3. Josefine's work plan
3.1 Validate and compare with Oasis3 the following interpolations:
· nearest-neighbour 2D
· bilinear (Note: Arne Biastoch for IFM_GEomar checked this interpolation from grid Reglonlatvrt to grid irrlonlat_regvrt and compared with Oasis3)
·
bicubic (Note:
Arne Biastoch
for IFM_GEomar checked this interpolation from grid from Reglonlatvrt
to grid
irrlonlat_regvrt and compared with Oasis3; comparison with results
obtained by
LEGI at
for the following grids (with "none" for the vertical interpolation):
· reglonlatvrt
· irrlonlat_regvrt
· gaussreduced_regvrt
· irrlonlatvrt
3.2 Validate the following interpolations:
· nearest-neighbour 3D
· trilinear
for the same grids.
For now, these checks will be done mainly without partitioning; they will be redone with partitioning after global search is implemented. One or two checks could be done now with partitioning to detect problems not linked with the global search.
3.3 Adaptation of proto_ex to the SRE; add a Gaussian reduced grid in one model; proto_ex is the only example that should be released to the community; the other examples are for development tests.
3.4 Start thinking about 2D conservative remapping implementation (global search).
3.5
Implement
vertical interpolation (linear and nearest-neighbour)
4. Global search
·
Work in progress;
should be
done for end of February. Action: Hubert.
5. Future implementation of conservative remapping
· One target cell will belong only to one EPIOt; for each target cell, the PSMILe will have to identified all the participating source cells, even coming from different partitions (see conservative_remapping.pdf)
· Uwe Schultzweida from MPI-M (currently working on NEMO+Echam5) can be asked for help, if there is a well-defined task.
6. Search algorithm (see search_algorithm.pdf):
·
We should add a
new key word
<if_outside> (below
<(bi)linear>, (bi)cubic>, <nneighbour(2d)>, and
conservativ(2d)
key word ; same level than <if_masked>) so to force the use
of the
nearest-neighbour value for the two following cases : 1- target
point
which mesh overlaps at least one source mesh but which does not fall
itself
into any source grid mesh (case A), 2- target point which mesh does not
overlap
any source mesh and which does not fall itself into any source grid
mesh (case
B) (the point would have to be added to one existing EPIOt). Action :
Sophie, Rene.
·
Action :
Rene to verify effect of keyword <if_masked> for case A.
7. Periodic (self-overlapping) grids:
· PSMILe behaviour in case of self-overlapping grids is OK; should be treated with special care when conservative remapping will be implemented.
8. Corner definition
· The longitudes of the corners of one cell have to define the size of cell (e.g. a cell with corners at 5 and 355 is a cell of 350 degrees), but there is no particular restriction in the numbers used (e.g. numbers greater than 360, or negative numbers are OK). Action: Sophie to clarify this issue in the documentation.
· SCRIP library in Oasis3 does not support cells larger than 180 degrees
· Warning is always printed by the PSMILe when cells larger than 180 degrees are detected; however, automatic correction of those cells is not desirable.
·
For the
definition of the ORCA
grid corners, LOCEAN people should provide proper corner definition for
ocean
cells following rule above for the different resolutions (this is not
the case
today with the current definition for ORCA2). Action: Sophie to add some
comments on the NEMO wiki page.
·
Corner ordering:
The rule
described on the Oasis4 development page is valid: Fortran ordering if
applicable; counterclockwise looking in direction of z at each level
for
irrlonlat_regvrt, irrlonlat_sigmavrt, reglonlat_sigmavrt,
unstructlonlat_regvrt, unstructlonlat_sigmavrt; volume_type needed for
irrlonlatvrt and unstuctlonlatvrt.
Action: Sophie to
clarify this issue in the
documentation and specify which grid type are currently supported.
· GEMS project (J. Flemming) 0-meridian/date line problem: We still do not know what exactly the problem is.
9. Status of Gaussian reduced grid support
· On one process, all currently interpolations (bilinear, bicubic with 16, nearest-neighbour) are supported.
·
With a
partitioned reduced grid
in I and in IJ, there are still some problems with high number of
partitions. Action: Rene
to check this problem.
10. Status of vertical interpolation "none" for the different interpolations
· With identical vertical grid, there is no problem; the exact point is found. If the vertical grids are not identical, the source point is not correctly identified in all cases. This problem was detected for the simplemggred example because vertical levels do not match (but is not a problem for IFS-CTM coupling). The corner locations need to be used. Action: Rene to check this problem.
7)
Repartitioning
of gridless grids (GEMS project)
· 3 exchange configurations are possible (See gridless_grid.ppt). Configuration 2 is recommended.
8)
SMIOC
information
·
We should
probably not repeat
the description information in the SMIOC (the 3 beta users we had up to
now
were confused) and modify PSMILe accordingly. Action:
Sophie.
3)
Current
development platforms
· OASIS4 tested and run with toy examples on:
· Intel Pentium 4 Workstation Cluster
· SGI O3000/2000 server with MIPS 4 processors and IRIX64
· SGI IA64 Linux server Altix 3000
· NEC SX6
· AMD 2800 Cluster
· IBM Power 4
· We should set-up a wiki page where we could keep ourselves updated on this. Action: Sophie.
4)
Status
of interactions with external project:
· SMHI: use of Oasis4 with circumpolar grid causing problems under work.
· IFM_Geomar: Echam5-NEMO with Oasis3 (with root process only because of misunderstanding about Oasis3 possibility) operational within weeks; will expand to Oasis4 (for end of march 2006, probably as test-bed in the first phase)
·
MPI: Same
coupling interface
for Echam5-MPIOM than Echam5-NEMO at
5)
Workshops/meetings
to come
· Ateliers Modélisation de l’Atmosphère, Météo-France, Toulouse, 01/2006 : poster by Sophie; will be reused at EGU.
· EGU 2006:
· 2005 (joined) session was too technical. For 2006, the topic has been extended to have our own session (not a PSI focussed-session, but the organisation can still be mentioned as PSI activity).
· Two or even three abstracts can be submitted from our side: 1- PSI in general, 2- coupler in general, 3- search library in particular.
· Kyosei workshop
· CCRLE asked to present their activities by their managers: search library with links to IFM-Geomar application will be presented.
6)
Additional questions:
The
following questions were discussed:
·
prism_set_subgrid:
there is no prism_set_subgrid_1d_real.F90 ? As it is not supported
yet,
prism_set_subgrid will be removed from the official distribution.
·
prism_set_angle:
argument `new_angle'
is missing? Same remark than above.
·
prism_def_var:
for subgrids, var_nodims(2)
can be the number of subgrids (even if it is already given in
prism\_set\_subgrid) or 0; for vectors, would var_nodims(2) can be the
number
of components (always 3) or 0.
·
prism_def_grid:
Gridless grids
must always be defined 3 numerical dimensions (possibly with 1 and 1 as
2nd
and 3rd dimension.)
·
Juha Vierinen’s
question:
“In the case of a aerosol model and atmosphere model, the coupling is
3D
and there are about 5-6 parameters (or transients). The only practical
way to
do this is by attaching the aerosol model directly to the atmosphere
model to
ensure that these
parameters are found within the main memory of the same processor. I
understand
that it is not possible to use the OASIS4 interpolation
facilities in this situation at all (because the atmosphere and aerosol
model
in the same process). Would it be possible to create some addition
to the PSMILe specification that makes this possible?”
We decided to keep a separate Transformer so to have a buffer in the
communication. Furthermore, if all MPI exchanges are effectively done
on the same node and if MPI is well implemented on the platform, the
MPI exchanges should be equivalent to a memory copy. So the MPI
exchanges can be (hopefully) not a too big overhead, even with a lot of
communication; this depend on the implementation of course.
7)
Strategy for future
developments
·
See
oasis4_developments_030106.html
· We could consider submitting a general paper on Oasis4 to the “Journal of Atmosphere and Ocean Technology”. Once scientific results are obtained (e.g. in collaboration with Noel and Wonsung for the Echam5-NEMO coupling), we could consider submitting another paper in a scientific journal. Action: Sophie, Rene.