cig_logo.png

CIG Dynamo Benchmarking Workshop

NCAR, Boulder, February 2015 


Meeting Goals: 

I. Present the results of the CIG community dynamo benchmark exercise. From the results of this community exercise, seek to determine how best to develop state-of-the-art community dynamo codes and numerical/theoretical tools that allow for more realistic simulations of dynamo action under core-like conditions. 

II. Start community-wide discussion of INCITE project and the numerical simulations to be carried out. 

I. Benchmark results

  1. Typically, 2D decompositions scaled up to 4096 cores, the max available in this study
  2. Spectrally accurate codes perform well in simplistic benchmark tests, such as the ones carried out here
  3. For best scaling codes (Calypso, LSD, RaC, SFEMaNS), we need to extend tests to max cores on Stampede (~16k cores),
  4. Port codes to Mira and see how they scale up to 100k cores and beyond
    1. Repeat standard benchmark tests
    2. Weak scaling tests with problem complexity increasing in proportion to resolution

II. INCITE Award 

   1.  Year 1
         i.      Optimize Rayleigh Code
         ii.     Carry out Weak scaling performance tests on Mira/CETUS up to 100k+ cores
         iii.    Survey / performance testing of all three cases (Jovian, Solar, Geo)
         iv.    Jovian Case: Simultaneous Shallow vortices-Jets/Deep jets/Dynamo

   2.  Year 2
         i.     Finish Jovian Hero Case
         ii.    Solar hero: heat transfer in multi-scale compressible convection  

   3. Year 3
         i.      Finish Solar Hero Case
         ii.     Geodynamo case: large-scale structures in turbulent metal dynamos

   4.  Output data:
         i.      Time series
         ii.     CMB output file
         iii.    Axisymmetric
         iv.    Spectral Information:
         v.    Checkpointing

Sign In