[MITgcm-support] problems at tile bdys?

Jody Klymak jklymak at uvic.ca
Mon Jun 17 01:00:06 EDT 2013


Hi Jean-Michel,

Thanks for your willingness to look at this - I'll put together a test-suite in the next couple of days. As often happens, maybe I'll find the error!

One thing that strikes me is that my 128-core simulation was using 47x16-sized tiles in a 3008x32 domain (i.e. 64x2 tiles).  My 64-core simulation, which is running fine right now, is 94x16 (same domain, 32x2 tiles).  Could it be that there is an even/odd problem?  I imagine most of the time folks do things in powers of 2, or at least with even-sized tiles.  

Anyway, happy to send you the link to a test case, but it may be a couple of days.

Thanks a lot,   Jody

PS, I'm using c63o.



On Jun 16, 2013, at  21:35 PM, Yuan Lian <lian at ashimaresearch.com> wrote:

> Hi Jean-Michel,
> 
> I found the similar behavior of MITgcm when there are multiple tiles in x direction. The issue is particularly obvious when active tracers are enabled, e.g. water cycle, where I can see the periodical oscillation of tracers including temperature at the boundaries between the tiles. The problem is less apparent in y direction. I am wondering if it has something to do with exchange routines in x direction. The simulations are done in 3d channel. Thanks.
> 
> Best,
> Yuan 
> 
> Jean-Michel Campin <jmc at ocean.mit.edu> wrote:
> Hi Jody,
> 
> Could you send (in a tar file) the directory that contains all the 
> modified/customized files you are using (the argument of "genmake2 -mods" command)
> as well as all the parameter files (all data* files) for this set-up ?
> Also, if you are not using the latest version of MITgcm code, which version 
> (checkpointXYz ? or time of the latest "cvs update") are you using ?
> 
> It's possible that we can find the problem like this; otherwise, will have 
> to try to reproduce it (and in this case will also need the binary input files).
> 
> Cheers,
> Jean-Michel
> 
> On Sat, Jun 15, 2013 at 11:03:30PM -0700, Jody Klymak wrote:
> 
> Hi All,
> 
> I ran a simulation in a 2-D channel but three-D in setup, and
> all ran fine.  Well except it stayed 2-D because there was nothing to cause there to be spanwise motion.  The simulations are nonhydrostatic.  
> 
> So, I added a very small amount of three d by a few different methods, and everytime, no matter how I set up the three dimensionality (bottom roughness, initial velocities, surface displacements) there would an instability at the same spot in x.  It turns out it is a tile boundary in the middle of the domain.  
> 
> I'm using advection scheme 33, and chnaged OLx and OLy to 5, but still the same problem.  I tried a few other advection schemes, with the same issue (it happens in a hundred or so timesteps, so its easy to debug, at least).  I also tried upping the iterations on the CG3d solver, but that didn't help at all.
> 
> I've now gone back to the same setup I started with, but changed by tile layout (half as many tiles), and the instability is gone.  
> 
> Has anybody run into this?  Any suggestions for a fix
> or what to look for?  (I guess the stupid fix is not to use that tile configuration, but that seems flakey).
> 
> Thanks,   Jody
> 
> 
> --
> Jody Klymak    
> http://web.uvic.ca/~jklymak/
> 
> 
> 
> 
> 
> 
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> -- 
> Sent from my Android phone with K-9 Mail. Please excuse my brevity.
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support

--
Jody Klymak    
http://web.uvic.ca/~jklymak/




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20130616/1e687167/attachment-0001.htm>


More information about the MITgcm-support mailing list