minutes_wp3a

OASIS4 Developments and bug fix priorities
(S. Valcke - October 29, 2007)

Bug under consideration High priority developments Medium priority developments Low priority developments Bug solved

Bug under consideration

SCRIP problems near the North Pole (in OASIS3 and OASIS4)

MetOffice (see Richard's mail Sept. 13 2007)
playback example (see Rene's mail Sept 04 2007)
BT42 to ORCA
for regular grids (Sophie, see Sophie's mail June 07 2007)
priority: high

Interpolation for ORCA grid (see also Kristian's problem 3; 09/14/2006)

status:

Laure tested ORCA to ORCA exchange with Sebastien's new corner definition: ORCAT-U-V all using the same (ORCAT corners) are OK.
Check if associating U points to to the cells formed by the V points or to the cells formed by the F points does not change the results (Laure)
2d conservative:

Overlapping problem: pb for i=181 and 182 is solved (by excluding them from the valid shape); pb for j=149 and 1/2 of 148 to check (excluding them from valid_shape?); test with no mask for overlapping points (currently they are masked).
Solved: Pb on the date line problem.

interpolation ORCA->ALADIN: some non-masked target points do not receive any value (see Laure's mail June 06 2007 and René's answer on June 08 2007, and other mail exchanges June 18th); René to improve/correct the nearest-neighbour extra search on the ORCA grid.
Solved: Results are good except for two grid points in the Arctic ocean; Hubert is working on this
Solved: interpolation ORCA->ORCA: strange results- partial lines of target points with psmile_undef or partial ines of target point with initial value (see Laure's mail June 08 2007)

(solved) symptom: corner definition seem to be wrong (shift by 1 in i and j)
priority: high

Memory leak in the driver/transformer + pb in deallocation (see Rene's mail about prismtrs_deallocate 02/22/07) (Sophie)

status: Sophie to test with IBM compiler with long runs; check dla_trans_in)
priority: high

Pb of PSMILe-T synchronisation: (see Sophie's mail to Rene on August 21)

priority: medium

Pb with extra search (nneighbour) for ORCA->ALADIN grid (cf Laure's mail June 06 2007)

Points in England channel do not receive nearest-neighbour value but values from further points
Under consideration by René (see René's mail June 08 2007)
priority: medium

Transformer bug for interpolation on 3D degenerated grids, i.e. 2D (Sophie)

cf Rene's mail dated 03/07/2006: When 2d grids are specified but trilinear interpolation is chosen the weights that are calculated by the driver are wrong. Independent of the geographical source and target locations weights are all 0.5 (see also mail Transformer bug 08/25/20006)
status: one bug fix provided for prismtrs_trilinear_weight.F90; correct the number of neighbours transfered for target point located on the bottom (or top) face of first (or last) source cells (Rene)
priority: medium

Johannes' problem; interpolation at poles (René, Sophie)

symptom: ripples near the poles for interpolation from Gaus Red for bilinear and bicubic
temporary solution: introduce a 4th behaviour "mirror" which results in extension of the domain beyond the poles by mirroring the domain for the local search and adapt Transformer (René, Sophie)
status:

solved with CPP key __MIRROR_POINTS for bicubic
to implement for bilinear (high priority)
implement "new" mirror option (more physically based)

priority: medium

Spurious areas at the borders of the MPI domains of the target model

reported by Johannes (june 07), we asked Johannes if this problem still exist, he explained that the spurious areas appear on the target grid at the border of the source MPI domains so they are probably caused by the lack of global search on the source grid
status: should be solved with implementation of global searchGlobal search for irrlonlat_regvrt + reglonlatvrt (Hubert, René)

Treatment of periodicity in j (cf Rene's mail to Johannes on 23/02/2006)

quick fix by René in March to overcome this problem
full treatment later with low priority

Vertical bounds in debug files for regvrt grids with 1 level only

priority: low

nneighbour2d /3d for reglonlatvrt and irrlonlat_regvrt grids

All cases work now, but they gives sligthly different results; investigate why (Sophie)
priority: low

High priority developments

<> Global search for gaussreduced_regvrt : to validate (Hubert, René)
2D conservative remapping -not parallel

implemented and tested for irrlonlat_regvrt, reglonlatvrt, gaussreduced_regvrt, ORCA (see "Bugs under consideration" above)

2D conservative remapping with global search (René)
Use of Trac (SV)

Interpolation quality check (parallel + monoproc) + comparison with OASIS3 results (Laure, Sophie)
Toymodels reproducing CICLE coupled models (parallel + monoproc) (Marie-Pierre)
Interpolation using a user pre-defined set of weights and addresses (Sophie)
IO for "gridless" grid (Reiner)
Distributed mode with netcdf, see Reiner's mail 06/12/2006 (Reiner)
implement if_masked for nneighbour2D/3D adding vneighbours -see comments in oasis4/doc/DescConf.tex (no if_missing nor if_outside for now) (Rene)
2D conservative remapping: Normalization by true area (Sophie)

Medium priority developments

Adaptation of PSMILe to read interpolation/communication information from files at restart instead of redoing all calculations at the beginning of each run.
Simple research algorithm applicable for all grids (monoproc, as in OASIS3)
Porting on CERFACS IBM (Sophie)
Porting on Meteo-France NEC (Laure, Marie-Pierre)
Review 2D1D interpolation (Sophie, Laure)
Clean coding of Transformer interpolation routines (check SX cpp key) (Sophie, Marie-Pierre)
Interpolation behaviour at domain boundaries (cf minutes of 25-27/10/2006 meeting in Sankt-Augustin)
Problem with lag greater than 1 coupling period (2 time occurence in the restart file, see mail from Josefine 05/10/2006)
Use of FCM for OASIS4 (and OASIS3) build system
Warning in psmile_def_domain for the 3rd direction for gaussreduced_regvrt grids (cf Laure's mail 03/22/2007)
Combination of source fields (for CICLE) (Sophie)
Change masks in T form integers to logical (Sophie)
Review XML files and PSMILe API (?) in relation with Curator and Metafor (in particular, remove all description information from SMIOC, change grid and grid family for resolution et grid, put nbr_overlap in PSMILe API) (Sophie)
Separate horizontal and vertical grid definition in API (? cf gridspec.nc)
Implement more SCC and SMIOC validity checks in Driver (Sophie):

no error message is produced if the source of a field is not defined properly in the smioc files
if no max_proc provided, use by default min_proc + number of procs (cf Philippe's problem)

Improve documentation on interpolation schemes (Sophie)
Provide database of users, FAQs, CPU statistics on the web/wiki (Sophie)
Ensure coherence with PRISM SVN server in Hamburg (Sophie)
Papers in "Earth Science Informatics" (ESI, Springer) or possibly Geoscientific Model Development (EGU)
More PSMILe function to get SCC and SMIOC info in model code (Sophie)
Support of bundles, subgrids, vectors, and bundle of vectors
Automatic nightly build and test system
Check time operation below the prism_get (IO data only) and related returned info code (Sophie)

Low priority developments

Transformer relocated in source PSMILe (Kristian's need to run one component sequentially one after the other - no load balancing pbs)
Check PSMILe behaviour for input from file: by default, time linear interpolation within the bounds?
prism_info routine (tu return the meaning of info arguments)
Remove need of SCC in standalone mode
Support conditional_computation in PMIOD (CM requirements) e.g. if an output transient is calculated only under a certain condition or after a certain period of time.Support more than one components running sequentially into one application
Support of GME grid
Test spawning of one application on different hosts
Additional Driver functions (prism_timestamp, prism_restartsaved, prism_message) ; provide more information on load balance (e.g. how long Oasis4 waited on a call)
Interpolation

Implement tricubic interpolation
3D conservativ3D (contact Balaji for method), user3d
1D cubic, conservativ1D, user1d
reconstruction of vertical position of grid points based on the units and auxiliary fields (see CF convention)
1D1D1D:

Can be used for grids with vertical units into which linear interpolation makes sense.
1D1D1D supported in PSMILe following user's choice in SMIOC ( inter1D,interp1D,interp1D)
1D: 1Dnneighbour, linear, none
No need of particular PS-T transfer routine

Driver: Check that interpolation method chosen is appropriate for the grid (e.g. check that vertical units are in meters if nneighbour3D or trilinear is chosen (SV)
PSMILe calendar tool (SV) on demand
Support of other type of exchange_date (once, precise, lastdayinmonth, list)

Support of source_time_operation: tmin and tmax

Support of reduction, coupling and I/O after reduction

Support of unstructured grid (PS I/O already OK if local partition can be expressed with only an offset in the global index space. For unstructured grids, this is the case if the local partition covers one segment in a 1D global space), expression of connectivity.

Volume type: for irrlonlatvrt and unstuctlonlatvrt grids, a description of the way the corners are connected is needed. An additional PSMILe call defining the connectivity between the corners could be used, but the prefered solution for now would be to explicitely defined different volume types in a User's Guide and refer to those explicitely in the SMIOC.

Saving of coupling restarts more frequently than run chuncks

Treatment of "abort restart"

Support of grid evolving with time

Automatic verification of what has changed and recalculation only for what has changed?
Problem if the grid changes between prism_put and coreesponding prism_get with lag
Problem of automatic writing and reading of coupling restarts

PSMILe C API

prism_def_var: arguments redundant with SMIOC information (SV)

PSMILe I/O:

Interaction with MetOffice; asynchronous IO server, netCDF vs HDF performance, etc.
Add option in SMIOC and use it in PS so that grid information is not written to the file with the data (RV, SV)
Format supported: mpp_ieee32 OK for I; mpp_native for I; mpp_ascii; GRIB; other
Output file size chosen by user; add a keyword in SMIOC (RV, SV)

Bug solved and developments done

Global search for irrlonlat_regvrt + reglonlatvrt + irrlonlat vrt: OK for all interpolation except 2d conservative remapping (Hubert, René)
Interpolation problems with Gaussian Reduced as source grid:

Bilinear form BT42 to AT42 with if_masked=nneighbour: it aborts; solved by checkin from 21 Aug 2007 (Rev 1480)
Bilinear/bicubic with if_masked=nneighbour BT42 to ORCA2: some points are still at psmile_undef; solved by Rev 1487
2D conservative: pb of transfer between PSMILe and Transformer (see Sophie's mail to Rene on August 20); solved by Rev 1530.
priority: high

Use of Subversion

Interpolation between the grids NEMO/ORCA2 T and ALADIN which are both irrlonlat_regvrt (in the frame of CICLE), when if_masked=nneighbour.

Symptom: aborts in psmile_mg_srch_nneigh_irr_dble.F90: line 574 (see Laure's mail, 05/29/2007)
Solved by René (01 june 2007)

Interpolation between two gaussian reduced grids when the target grid has a mask (cf mail by Laure on May 14th 2007)

Symptom: Aborts in psmile_mg_found_loc_to_3d.F90

Solevd by René (as well as other problem detected when interpolating between two different gaussian reduced grids)

nneighbour2d for reglonlatvrt (René)

symptom: was not supported by the PSMIle (see René's mail 01/12/2007)

now implemented and check OK by Sophie (May 14, 2007)

nneighbour2d /3d for irrlonlat-regvrt (René):

symptom: only the the target points that coincide with the source points receive something (see René's mail 01/12/2007)
now implemented; checked by Sophie (May 14, 2007)

__PSMILE_IO_SYNC directive in mpp_io (this invokes calls to mpp_flush.) (Reiner)
Memory leak in SHMI (René)

Treatment of periodicity in j (cf Rene's mail to Johannes on 23/02/2006)

symptom: memory increases with calls to PSMILe_bsend and terminates in MPI_Isend.
status: maybe solved with new psmile_bsend.c; confirmed by SMHI.

Ian Henderson's bug:

symptom: The target field has some zero values when the longitude of the grid are defined with numbers more negative than -180 (see Ian's mail dated 11/15/2006)
solved and checked in on trunk and development branch 01/18/2007

Richard Hill's problem (09/04): grid points at +- 90 (René)

status: solved and checked-in; confirmed by Richard.

Gauss reduced grid to high res ocean grid (Kristian's problem 2; 09/14 and 09/06) (René)

symptom: problems along the greenwich meridian
Johannes reproduced this problem in his IFS-OASIS4-CTM environment
status before 09/27: problem identified, idea for a solution exist: perform some operations now done in psmile_gauss_setup_dble/real somewhat earlier before the call to psmile_get_grid_extent_dble/real and already use the auxillary grid there instead of the true gauss grid
status 09/27: solved and checked in

Reg to Reg problem when no psmile_set_mask is called (Kristian's problem 1; 09/14) (René)

status: solved and checked in
happens only when there are matching grid points and non-matching grid points (see René's mail 09/22/2006)
non-trivial to fix if we want to maintain to optimisation done for reglonlatvrt grids
the problem can be worked around by calling a prism_set_mask

Date line problem:

reported by Johannes (june 07)
status: solved; Johannes confirmed in mail dated 30/10/2006