****** model is blowing up *********

[scs_wy@yahoo_cn]

Hi everyone
I am a CCSM user and I can run T42_gx1v3,T31_gx3v5 and T85_gx1v3 successfully on my cluster.
It is strange that when I run T42_gx3v5 there is a segment fault happened and it like something wrong about cam.
--------------------------------------------------------------
[node11:04764] *** Process received signal ***
[node11:04764] Signal: Segmentation fault (11)
[node11:04764] Signal code: Address not mapped (1)
[node11:04764] Failing at address: 0x16245528
[node11:04767] *** Process received signal ***
[node11:04767] Signal: Segmentation fault (11)
[node11:04767] Signal code: Address not mapped (1)
[node11:04767] Failing at address: 0x16819f68
[node11:04763] *** Process received signal ***
[node11:04768] *** Process received signal ***
[node11:04763] Signal: Segmentation fault (11)
[node11:04763] Signal code: Address not mapped (1)
[node11:04763] Failing at address: 0x16844528
[node11:04768] Signal: Segmentation fault (11)
[node11:04768] Signal code: Address not mapped (1)
[node11:04768] Failing at address: 0x1680ff68
[node11:04764] [ 0] /lib64/libpthread.so.0 [0x2ad997771c00]
[node11:04764] [ 1] /dcfs2/users/wy/case_0204_35/exe/case_0204_35/all/cam(sphdep_+0xc14) [0x6e7f14]
[node11:04764] *** End of error message ***
[node11:04766] *** Process received signal ***
...
------------------------------------------------------------------------
The file atm.log show that：
------------------------------------------------------------------------
...
SPHDEP: ****** MODEL IS BLOWING UP *********


Parcel associated with longitude 3, level 0 and latitude 1 is outside the model domain.
ENDRUN: called without a message string
----------------------------------------------------------------------
I don'g kown why,anyone can give some advice??
Thanks......................

 

[olson]

You are encountering a violation of the CFL condition. Somewhere the winds are too
strong for the timestep you are using. The default is 1200 seconds. Try turning on
the divergence damper for one day by setting the CAM namelist variable:
DIVDAMPN = 1.

This will damp any initial shocks the model may be encountering. The divergence damper will
shut off after 1 model day.

If that does not work, try using a smaller timestep such as: DTIME = 900

 

[scs_wy@yahoo_cn]

Thank you for your reply olson.
I didn't modify anything about CCSM3 script.Why some resolutions(T42_gx1v3 etc) could be run and some(T42_gx3v5) couldn't.
Is it my cluster calculation of control accuracy problem??
:confused:

 

[olson]

The 1200 second timestep for T42 pushes the limits of the model's ability to maintain stability. So
sometimes you may hit a scenario where the model goes unstable simply because the winds get just
a *little* bit too strong and tip the CFL condition to > 1. In fact, extra diffusion was put into the
model just so T42 could be run with a 1200 s timestep.

You didn't mention if this happens immediately or if it happens after many months or years of
model runtime? I do not know if it would have anything to do with your cluster
configuration - it's possible.

 

[scs_wy@yahoo_cn]

Hi olson
I modify the cam namelist variable DIVDAMPN and DTIME respectively.Unluckily the same bolwing problem still happen.Could it be that there is any other reason for this blowing problem ???

 

[olson]

Yes there could be but I would need more details to diagnose the problem.

 

[scs_wy@yahoo_cn]

Could you give me you email and I will send you all log file
Thank you