Known Problems in Version 3.0.1.1 (Details)
Compile Optimization Issue on IBM Power Series (posted 8/05/08 - fixed in version 3.0.1.1)
Issue: The default optimization level for the WRF compilation on IBM Power Series is -O3 -qhot. This level is considered an aggressive optimization. Compared with -O2 (which was used in WRF V2), the -O3 -qhot optimization level gives an approximate 30% performance speed-up. It is known that the cu_physics=99 option is especially impacted with the -O3 -qhot optimization level.
Solution: If you have a concerns about the WRF default optimization level, you may lower this in the configure.wrf file to either -O3 (without -qhot) or further reduce it to -O2.
Maximum number of processors for given domain sizes (posted 8/05/08 - fixed in version 3.0.1.1)
Issue: The ARW core requires that no patch (subdomain allocated to an MPI task) be no smaller in its horizontal dimensions than 6x6 for coarse domain and no smaller than 9x9 for a nest. For each dimension of each of your domains, divide the domain width by the number of tasks over which that domain will be decomposed. The result should not be less than 6 for the coarse domain and not less than 9 for any nest. For example, if you are running a simulation with a nested domain that is 424 in X and 325 in Y, then the largest number of tasks in X is 47 and in Y is 36 (for a total of 1692). The smallest number in X and Y over all domains is the most tasks you can run for the simulation.
The correct number of processors in X and Y might not be the number in X and Y that the model computes automatically. For example, running with 16 MPI tasks, the model will come up with 4 processors in X and 4 in Y on its own. If you need 4 in X and 2 in Y, you must override the model's choice by specifying nproc_x and nproc_y in the domains section of the namelist.input file.
* Error behavior:
Beginning with version 3.0.1, the model will generate a message:
tfp_message: module_dm.b <line number>
and abort. Prior to 3.0.1, the behavior is more dangerous: the code may still run, but incorrectly and with the possibility of crashing unexpectedly later in the run.
Also beginning with 3.0.1, there is an option that will cause the model to detect this "overdecomposition" condition and then automatically reduce the number of processors decomposing a problem domain. To enable this option, add -DALLOW_OVERDECOMP to the ARCH_LOCAL line of your configure.wrf file, clean, and recompile. Note that the -DALLOW_OVERDECOMP option is new and not yet fully tested. Use at your own risk. For the time being, if you still run into trouble, work out the maximum number of processors by hand using the method described above and only specify configurations that will not cause an overdecomposition condition.
netcdf 4 (posted 8/05/08)
Problem: WRF does not yet support netcdf 4. Please use netcdf 3 instead, and 3.6+ if you have file sizes greater than 2 Gb.
Occasional Spurious High PBL Height In MYJ Scheme (posted 8/05/08)
Problem: MYJ PBL scheme occasionally produces unrealistically high PBL and high TKE in sporadic columns. This is reported by PSU researchers.
Solution: Download the modified module_bl_myjpbl.F (from PSU). The scheme computes the PBL depth by using the TKE profile criteria from Penn State, instead of the current MYJ method that searches for the first level of background TKE. This fix reduces the PBL height in general.
OpenMP compile in V3 (posted 5/16/08, updated 8/05/08)
Problem: Certain physics options do not work correctly with OpenMP compile. These are cu_physics = 3 (Grell-Devenyi) and 5 (new Grell 3D scheme); and ra_lw_physics = 3, ra_sw_physics = 3 (both CAM radiation). There is no known problem with these physics options if OpenMP is not used. Likewise, there is no known OpenMP problems with other physics options.
Solution: Do not attempt to use the above listed options with OpenMP compile.
Using Shared Libraries with NETCDF 3.6.2 and Later (posted 6/20/08)
Problem: If one encounters the following problem at runtime:
>wrf.exe
wrf.exe: error while loading shared libraries: libnetcdf.so.4: cannot open shared object file: No such file or directory
It probably means that you are using netcdf 3.6.2 with shared library build enabled.
Solution: For PGI compiler 6.1 or later, one can add -R/usr/local/netcdf/lib to LIB_EXTERNAL macro in the configure.wrf file, so the line looks like:
LIB_EXTERNAL = $(WRF_SRC_ROOT_DIR)/external/io_netcdf/libwrfio_nf.a \
-L/usr/local/netcdf/lib -lnetcdf -R/usr/local/netcdf/lib
Or, one can rebuild netcdf without shared library option enabled.
Same fix should be applied to WPS configure file, configure.wps.