[CIG-LONG] rifting test case

[Walter Landry]

Garrett Ito <gito at hawaii.edu> wrote:
> Hi Eunseo,
> 
>   My job was able to complete 2.2e4 yr in 36 hrs on 45 processors (the machine
> I used was http://www.mhpcc.hpc.mil/doc/jaws.html) .  Thus at this point I
> cannot tell the difference between Gale and SNAC for this particular case.  It
> would be nice to compare the codes more closely.  Can you run your job on say
> 45 processors at CIG with an good enough size timestep to optimize runtime and
> solution accuracy?   Note, I created a small box in which cohesion was set to
> zero to initiate the faulting.  It is located at x=27-28 km in the uppermost 2
> km.  I'm actually quite interested in the results because I find FLAC (and
> presumably SNAC) a much simpler code to understand and work with.
> 
> Walter,
>   If Ensuo can test SNAC at CIG would you mind running the attached input file
> on the same nodes?  It looks like this is actually a tough problem for Gale to
> solve as it took longer than cases I ran before and runtime is pretty
> sensitive to initial conditions.  This might be a good benchmark case for
> rifting problems and is what I will send to Steve to test on Underworld.

I ran the input file with 48 processors on Pangu.  It got to
TimeStep=4 (1.24e4 years) after 22 hours.  Running for another 26
hours and it did not finish another step.  Looking in detail, it seems
the linear solves get slower and slower.

So now I am very interested in how SNAC handles this problem once the
material really starts yielding.

Cheers,
Walter Landry
walter at geodynamics.org