"wrong choice of domain..."when testing upwelling

[Jin Qihua]

Dear all:
I'm a new user of ROMS, and I installed ROMS2.2 on IBM1600 cluster. I am tring to run the upwelling case. After the compile, I do get an oceanM executable, then the following job script is submitted:

#!/bin/ksh
#@ output = ./Output/$(Executable).$(jobid).out
#@ error = ./Error/$(Executable).$(jobid).err
#@ environment = MP_SHARED_MEMORY=yes; COPY_ALL
#@ notification = error
#@ network.mpi = sn_single, shared, US
#@ node_usage = not_shared
#@ wall_clock_limit=8:00:00
#@ job_type = parallel
#@ node =1
#@ tasks_per_node=8
#@ class= normald
#@ queue

./oceanM ocean_upw.in

But when I look at the output, it says wrong choice of domain 001 partition or number of parallel threads. I changed NtileI=1, Ntilej=8 in ocean_upw.in.

Can any of you tell me why?
Thanks in anticipation.
Jin

[jcwarner]

I am not familair with that cluster type, but ...

1) try submitting with just 1 processor. So set NtileI=1, NtileJ=1, nprocs =1

2) do you need to say something like
mpiexec -np 1 ./oceanM ocean_upw.in

or similar syntax??

[kate]

Jin,

I also use an IBM with a similar job file. I put poe in front of oceanM, though I don't know if it's necessary in your case:

Code: Select all

time poe ./oceanM External/ocean_nep.in

You might try this line before the ROMS line:

Code: Select all

poe hostname

This will give one answer per process so you can see if it really gave you eight processes. I assume you've checked the system you use and know that sn_single is the right thing on the network.mpi line. I use sn_all on our system, but I often ask for more than one node.