[aspect-devel] Parameter files/setups for scaling tests

[Rene Gassmoeller]

Hi John,

I think the setup is a good one for mantle convection models. We might
want to think about a model with plasticity (or at least nonlinear
rheology) at some point, but it is somewhat tricky to set up so that
different resolutions require the same amount of iterations (for weak
scaling tests). We might want to measure the time per iteration instead
of time per timestep in these cases (but that would not capture the
scaling of our preconditioner with model size).

I also think it is a good idea to test the scaling of the cheap solver,
since I expect that it might be somewhat different from the expensive
solver (less AMG solver steps for the A block, which are the main
limiting factor for the scaling). I am happy to repeat my performance
tests with the cheap solver on the systems I have tested so far.

Particle scaling tests can wait a bit. I have done some for the particle
paper (https://arxiv.org/abs/1612.03369), but I do not expect that large
of a difference between clusters, because the communication is not a
dominant factor (the algorithm and data structure is more important, but
that is not system dependent).


Cheers,

Rene


On 06/20/2017 03:01 PM, John Naliboff wrote:
> Hi all,
>
> Following the discussion on GitHub,  a few questions regarding
> performing additional scaling tests.
>  
> (https://github.com/geodynamics/aspect/pull/1552#issuecomment-304103931).
>
> There is one input file one the performance benchmark folder Rene
> setup (https://github.com/gassmoeller/aspect-performance-statistics).
> For additional clusters, should we use this setup for the strong and
> weak scaling tests? I'm planning to use this setup on COMET and
> STAMPEDE2, but replace the expensive Stokes solver steps with cheap
> ones. Does this sound good or is there a preferable alternate setup?
>
> I was not planning to do particle scaling tests right away, but happy
> to do so after running the initial scaling tests. 
>
> Cheers,
> John
>
> *************************************************
> John Naliboff
> Assistant Project Scientist, CIG
> Earth & Planetary Sciences Dept., UC Davis
>
>
>
>
>
>
>
>
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel

-- 
Rene Gassmoeller
http://www.math.colostate.edu/~gassmoel/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20170621/545e72f3/attachment.html>