Running CAM5 Standalone in Serial

[lneef@geomar_de]

I'm trying to run CAM5 standalone, but in serial mode rather than parallel. I compiled it after setting MPI_SERIAL = TRUE in env_conf.xml, but when I try to run it as a serial job (using Load Leveler), I get a core dump right after it executes the model.

I'm still trying out various settings in the load leveler options but I thought I'd throw this out there -- has anyone else tried to do a serial run and run into similar issues?

 

[eaton]

You can easily do this using a cam standalone script (example scripts are in models/atm/cam/bld/). Just give configure the -nospmd -nosmp arguments. I'm not sure what the procedure is from the CESM scripts via the F compset. I thought it was possible since we do supply an mpi-serial library, but I don't know the details.

 

[lneef@geomar_de]

Ok, thanks for the quick feedback! Actually compiling from standalone CAM scripts made me problems worse -- I run into a host of fortran compiler issues because the makefile generated by CAM's (standalone) configure program doesn't know the details of the machine on which I am running. I spent 2 days with a colleague trying to figure out which fortran flags need to be added to the make file to make CAM compile, but to no avail -- no matter what we do, there is always some mysterious complaint about the code.

For a standard CESM build, we just skirt this issue because we define the machine when we build the case, and then ./configure -case reads the macros file that is defined for our machine. What I don't understand is how to configure CAM alone (using the cam scripts) and still get it to read this Macros file, or at least get it to accept the settings for our system. [FYI, I'm running on the AIX machine Blizzard at the DKRZ in Hamburg, Germany.]

So in summary I'm running into two separate problems, and now sure which is better to tackle

(1) building CAM standalone using the standard CESM run scripts fails if I try to set MPI_SERIAL = TRUE.

(2) building CAM using its own build scripts fails because I don't know how to get the fortran settings of the machine I'm using to be "read" during the configuration.

 

[eaton]

I would guess that your AIX machine would only require small modifications from what we do at NCAR. You are correct that the CAM standalone scripts don't interact with the CESM Macros files. But CAM's configure command has several options to allow customizing the compile or link commands. And of course there's always the option of editing the Makefile.in template file if necessary (in models/atm/cam/bld). If you post the specific errors you're getting then I can make suggestions.