Curious problem with PE layer

[11438023@zju_edu_cn]

Hi there,I'm a newer to CESM. I met problems when test B1850CN compset with f45_g37 res. The problems are with PE layer configuration. 1, For the compset B1850CN, I tested two cases, (1) the first one running 12 months, and it spent 13 hours to complete (which seems good).(2) the second one I changed the “STOP_OPTION" to “ndays" and the “STOP_N" to “5”.  This time the model ran for 20 hours but still not complete.All other sets are the same.  I wonder if anyone know why the speeds differ so much? Is it because the cluster or my setting? 
2, I test the B1850CN case on our cluster with 128 tasks or cores (8 nodes, 16 CPUs).It reports error at the beginning of my run when I give the land part 32 tasks (in "env_mach_pes "32, 1, 96") and the others 96 tasks ("96,1,0"). But curiously when I change the land to 38 ("38, 1, 90") and the others 90 ("90, 1, 0"), it reports no error when running (still running). I don't know why
Considering before's discussion (http://bb.cgd.ucar.edu/cesm-122-segmentation-fault-when-running-model) , are these problems related with "DEBUG" or the POP2 models with intel mpi?    

 

[11438023@zju_edu_cn]

I solve these problems by changing the "DEBUG" in "env_run.xml" to "TRUE".  

 

[11438023@zju_edu_cn]

Here comes another problem.When i set the "DEBUG" as "TRUE", and all tasks in sequence, (all models the same in "env_mach_pes.xml" as "128,1,0"), it broke down at the beginning of my run. But when I turn the "DEBUG" to "FALSE", no error occured. This problem only occurs with all models tasks set in sequence. PE layer setting in #1 have no error at all!