re: built in tests on ibm-sp atm_stdio.nml

[jkmoore]

I have trouble running the built in tests using the ccsm3.0 release code. Tests pass with all tasks=1,
and with some models set =2. But
as total task numbers increase the model crashes with various error messages

0:ERROR: 0032-117 User pack or receive buffer is too small (0) in MPI_Waitall, task 0

or segmentation fault error or
ERROR: 0031-250 task 3: IOT/Abort trap

Looking at the logs this seems to happen when the atm model trys to read....
(shr_msg_chStdOut) read atm_stdio.nml, unit 6 connected to atm.log.041006-085913
(shr_msg_chStdIn) read atm_stdio.nml, unit 5 connected to atm.stdin

This is similar to the problem reported on the SGI on this bb. Someone at another school mentioned a similar problem also on an IBM-SP.

Has anyone found/solved similar problems?
thanks.

 

[gcarr@ucar_edu]

Your posting does not give me enough information to really be able to give a useful response. Often the MPI errors you are seeing indicate an MPI configuration issue. There are limitations on just how many cpus and threads can be given to each component. Try starting with all data models.