abangal2@ncsu_edu
New Member
I am studying CCSM's (ver 4.0) IO characteristics on Cray XT (JaguarPF @ ORNL) under different PE sizes.
PE sizes are automatically generated in *.run file based on the configuration in env_mach_pes.xml. Scaling beyond a certain point gives the following error message:
1. In general, how do I scale the problem so that I can run under large number of PEs (10,000 cores) ?
2. Different components have different PE sizes in env_mach_pes.xml. How do I scale these individual models ? Should they be increased proportional to their default size ?
PE sizes are automatically generated in *.run file based on the configuration in env_mach_pes.xml. Scaling beyond a certain point gives the following error message:
Code:
$ configure -case
cice decomp not set for tx0.1v2 on 10240 x 1 procs"1. In general, how do I scale the problem so that I can run under large number of PEs (10,000 cores) ?
2. Different components have different PE sizes in env_mach_pes.xml. How do I scale these individual models ? Should they be increased proportional to their default size ?