[CIG-SHORT] Uniform time-stepping for long runs on viscoelastic domain (Brad Aagaard)

[Brad Aagaard]

Michal,

The 95 files instead of 100 years is almost surely related to round off 
errors. If you want output every 1 year exactly, I generally use 0.9999 
years in setting the output time interval or set the time step skip to 9 
(0 means every time step) when using a dt of 0.1 years.

I agree that the long run hanging is suspicious if the simulation runs 
fine for fewer time steps. The best way to debug that type of situation 
is to use the debugger and turn on more output journals and PETSc 
monitors so you can see where the code is.

Regards,
Brad


On 6/7/15 12:00 PM, Michal Dichter wrote:
>
> Dear Brad,
>
>
> Thank you for your response. I do have some concerns about the PyLith
> "counter" deviating so far from the requested total time. Here is what I
> have observed for a uniform time-step of 0.1 years, requesting output to
> be written every 1.0 years:
>
>
> total_time = 5.0 years :: Code runs perfectly, outputs 5 files (one per
> year, as requested)
>
> total_time = 100.0 years :: Code runs to completion, but only outputs 95
> files (not 100, as requested)
>
> total_time = 500.0 years :: Code runs great for 99% of run and then gets
> stuck forever (forever = no progress after 24 hours)
>
>
> I am pretty sure that PyLith is not busy calculating state variables for
> the "long run" that gets stuck, as this process takes no more than a
> couple of seconds to minutes for the cases of total_time = 5.0 and 100.0
> years. The file sizes I'm dealing with are not large, a couple hundred MB.
>
>
> Can you offer any insight as to why the counter deviates so much? Is it
> possible that this is causing the 500-year long run to get stuck?
>
> If so, what can I do to address the problem? The fact that the 100-year
> run (the "intermediate" case between a total_time that runs perfectly
> and one that doesn't run at all) only outputs 95 files when I've
> requested 100 files is a smoking gun to me.
>
> Thanks again.
>
> Best,
>
>
> Michal
>
>
> ____________
> Michal Dichter
> Department of Physics and Astronomy
> University of New Mexico
> (860) 459-0956
>
>
> _______________________________________________
> CIG-SHORT mailing list
> CIG-SHORT at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-short
>