You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The configurations used in the dev/ufs-weather-model branch utilize the CMEPS mediator to provide importState fields from other model components at each ModelAdvance for WW3.
Currently, the mesh cap allocates and fills the fields required by WW3 for two time levels of these import fields---eg, CX0 and CXN. The "next" fields are simply set equal to the "current" fields for the purposes of making WW3 happy when it "time interpolates" the forcing fields in w3updtmd. However, since T0==TN, the end result after w3updtmd is that CX==CX0.
The concept of time-interpolating the forcing fields is meaningless for the configurations where CMEPS is providing updated importState fields at a single time. The allocation of CXN simply wastes memory, as does the allocation of arrays which are associated with the time-interpolation and which serve no purpose. For the ocean currents, this includes arrays CA0,CAI,CD0,CDI.
Depending on configuration, the total number of unnecessary arrays is 15, all of which are allocated as the global domain on each DE. Testing indicates these arrays can be safely allocated to size=1, the "time interpolation" in w3updtmd can be skipped and all model results are B4B in UFS. This reduces the memory footprint for the dev/ufs-weather-model branch. Initial testing has indicated some potential for also speeding the integration time. A PR is being prepared.
The text was updated successfully, but these errors were encountered:
The configurations used in the dev/ufs-weather-model branch utilize the CMEPS mediator to provide importState fields from other model components at each ModelAdvance for WW3.
Currently, the mesh cap allocates and fills the fields required by WW3 for two time levels of these import fields---eg,
CX0
andCXN
. The "next" fields are simply set equal to the "current" fields for the purposes of making WW3 happy when it "time interpolates" the forcing fields inw3updtmd
. However, since T0==TN, the end result afterw3updtmd
is thatCX==CX0
.The concept of time-interpolating the forcing fields is meaningless for the configurations where CMEPS is providing updated importState fields at a single time. The allocation of
CXN
simply wastes memory, as does the allocation of arrays which are associated with the time-interpolation and which serve no purpose. For the ocean currents, this includes arraysCA0,CAI,CD0,CDI
.Depending on configuration, the total number of unnecessary arrays is 15, all of which are allocated as the global domain on each DE. Testing indicates these arrays can be safely allocated to size=1, the "time interpolation" in
w3updtmd
can be skipped and all model results are B4B in UFS. This reduces the memory footprint for the dev/ufs-weather-model branch. Initial testing has indicated some potential for also speeding the integration time. A PR is being prepared.The text was updated successfully, but these errors were encountered: