Differences in NTASKS affect the simulation result of snow

[zhangmeixin]

I use the CESM2.1.3 to simulate the snow of Tibetan Plateau. Two sets of identical experiments were conducted（FHIST, begin from 1979, 0.9x1.25), with the only difference being the nodes. Upon examining the namelist, the only difference was npr_yz. However, the results of the experiments differed significantly, especially in the interannual variability of snow-related variables.
The nodes show as follows:
One is:
./xmlchange NTASKS_CPL=-26,NTASKS_ATM=-26,NTASKS_LND=-24,NTASKS_ICE=-2,NTASKS_OCN=-1,NTASKS_ROF=-24,NTASKS_GLC=-1,NTASKS_WAV=-1,NTASKS_ESP=-1
./xmlchange ROOTPE_CPL=0,ROOTPE_ATM=0,ROOTPE_LND=0,ROOTPE_ICE=-24,ROOTPE_OCN=-26,ROOTPE_ROF=0,ROOTPE_GLC=0,ROOTPE_WAV=0,ROOTPE_ESP=0
The corresponding setting of npr_yz is 64,9,9,64

Another One is：
./xmlchange NTASKS_CPL=-18,NTASKS_ATM=-18,NTASKS_LND=-16,NTASKS_ICE=-2,NTASKS_OCN=-1,NTASKS_ROF=-16,NTASKS_GLC=-1,NTASKS_WAV=-1,NTASKS_ESP=-1
./xmlchange ROOTPE_CPL=0,ROOTPE_ATM=0,ROOTPE_LND=0,ROOTPE_ICE=-16,ROOTPE_OCN=-18,ROOTPE_ROF=0,ROOTPE_GLC=0,ROOTPE_WAV=0,ROOTPE_ESP=0
The corresponding setting of npr_yz is 64,13,13,64

The time series graph of the regional average snow water equivalent in the Tibetan Plateau area for the two experiments is as follows:

 

[nusbaume]

Hi zhangmeixin,

I believe answers are task independent only if the following XML change is made in the relevant case(s):

./xmlchange BFBFLAG=TRUE

If that doesn't work, then if you are running on an NCAR machine (e.g. derecho) can you point me to the paths of your two cases? We frequently test that changing the task count doesn't change answers, so hopefully something is just off somewhere.

If you are running on a different machine, then do you know if the code has been officially ported and validated? For example, has anyone done the steps described here:

7. Porting and validating CIME on a new platform — CIME master documentation

Note that you may have to change cheyenne to derecho in the example test commands.

Anyways, I hope that helps, and have a great day!

Jesse

 

[zhangmeixin]

Hi zhangmeixin,

I believe answers are task independent only if the following XML change is made in the relevant case(s):

./xmlchange BFBFLAG=TRUE

If that doesn't work, then if you are running on an NCAR machine (e.g. derecho) can you point me to the paths of your two cases? We frequently test that changing the task count doesn't change answers, so hopefully something is just off somewhere.

If you are running on a different machine, then do you know if the code has been officially ported and validated? For example, has anyone done the steps described here:

7. Porting and validating CIME on a new platform — CIME master documentation

Note that you may have to change cheyenne to derecho in the example test commands.

Anyways, I hope that helps, and have a great day!

Jesse

Hello！I am not running on an NCAR computer, and indeed, the porting has not been validated with this step. I will follow your suggestion and run a test first, thank you!