[aspect-devel] temperature assembly timing

Wolfgang Bangerth bangerth at math.tamu.edu
Thu Sep 19 04:31:38 PDT 2013


On 09/18/2013 02:44 PM, Thieulot, C. (Cedric) wrote:
> here are the timings in optimised mode : still spending 30% of the time in the
> temperature assembly.
>
>
> 16536 +---------------------------------------------+------------+------------+
> 16537 | Total wallclock time elapsed since start    |  1.21e+04s |            |
> 16538 |                                             |            |            |
> 16539 | Section                         | no. calls |  wall time | % of total |
> 16540 +---------------------------------+-----------+------------+------------+
> 16541 | Assemble Stokes system          |      1801 |       355s |       2.9% |
> 16542 | Assemble temperature system     |      1801 |  3.66e+03s |        30% |
> 16543 | Build Stokes preconditioner     |         1 |      3.62s |      0.03% |
> 16544 | Build temperature preconditioner|      1801 |       514s |       4.3% |
> 16545 | Solve Stokes system             |      1801 |  5.83e+03s |        48% |
> 16546 | Solve temperature system        |      1801 |  1.16e+03s |       9.6% |
> 16547 | Initialization                  |         2 |     0.339s |    0.0028% |
> 16548 | Postprocessing                  |      1801 |       261s |       2.2% |
> 16549 | Setup dof systems               |         1 |      2.05s |     0.017% |
> 16550 +---------------------------------+-----------+------------+------------+

This is relatively typical. Assembling the temperature system is expensive, 
because so many parts go into it (we need to evaluate the velocity, many 
coefficients, and compute the artificial viscosity which involves the second 
derivatives of the temperature field). We need to eventually look into whether 
it is possible to cut corners somewhere. That said, if you have models with 
spatially varying viscosity, or large processor counts -- in other words, the 
interesting cases -- then the Stokes solver becomes significantly more 
expensive and takes over much of the CPU time, so we've always felt that that 
is the more critical piece...

Best
  W.

-- 
------------------------------------------------------------------------
Wolfgang Bangerth               email:            bangerth at math.tamu.edu
                                 www: http://www.math.tamu.edu/~bangerth/



More information about the Aspect-devel mailing list