Parallel Computing: Few results


Benchmark shows that the time saved depends on the application we want to solve.

As the parallelization is more important on the linear systems solving, the most important gain comes on projects with large linear systems to solve. Best parallel efficiencies are then obtained on large 3D steady state projects.

The efficiency is even better when other operations than these linear systems solvings are limited. Speed-up factors are then weaker on transient projects which require a lot of non-linear operations and integrations.

On the most favourable cases, speed-up obtained can reach a factor 6 when using 16 cores. However, efficiency also depends on the resources used which must be adapted to the size of the project to solve. The following table summarizes some recommendations on which solver to use regarding the application and the size (in degrees of freedom) of the project to solve.


Future work on parallel computing for Flux will be on a global parallelization of Flux, allowing speed-up a bigger part of the solving process and increasing the total time saved.