MP_EAGER_LIMIT_LOCAL error

[bsf208@lehigh_edu]

Trying to run CESM 1.1 Yellowstone using default options for compiler and MPI. Builds fine and it seems to start up fine, then hangs, as seen in /run:

-rw-r--r-- 1 felzer ncar 524331 Feb 7 21:05 cesm.log.130207-195342


The run ends with the cesm.log file that includes the error:


ATTENTION: 0031-606 Unrecognized environment variable, MP_EAGER_LIMIT_LOCAL.


This error has occurred over several attempts to run the model. In startup 20-day load balancing tests the error did not appear to stop the run, however, in recent 50 year hybrid model run the model hangs and does not complete or produce history files.


Any thoughts on how to fix this?

 

[avenger@avenger_ws]

I see no MP_EAGER reference on both open-mpi nor mpich documentation, but on IBM's implementation of MPI there is one. Then this could have been input by someone using IBM's MPI library and (maybe) should not be present on CRAY's implementation of MPI (which, if same as here, is mpich2). For instance, the MP_EAGER_LIMIT_LOCAL is documented here: http://www.lrz.de/services/software/parallel/mpi/ibmmpi/It probably does nothing (may be present but is not used nevertheless) and is not exactly your error, but just warning. You should look for actual error messages for the cause your process dies.

 

[jedwards]

The MP_EAGER_LIMIT_LOCAL is a variable that was once used in IBM's mpi library but no longer is.   This message can safely be ignored.    The real error in this users run can be found further up in the referenced log file:  0:forrtl: No such file or directory
 0:forrtl: severe (29): file not found, unit 98, file /glade/scratch/felzer/F2000CNA_CTL/run/rpointer.drv

 

[avenger@avenger_ws]

I couldn't even guess in which machine would that log file be... :) Tried to help with what I had access to.