Weak scaling

Aug 7, 2017 00:00 · 112 words · 1 minute read

Good news that parallel solving in DOLFIN for quadrilateral and hexahedral mesh worked “out-of-the-box” with the changes made to FFC and FIAT. Now it is needed to see how well it performs.

Weak scaling test shows how the solution time varies with the number of processors for a fixed problem size per processor. I’ve adapted Chris’s demo https://bitbucket.org/chris_richardson/weak-scaling-demo for hex mesh.

Here are the results:

It is seen from the plot that assembly and defining function space takes roughly the same time when increasing number of processes. For solving hypre-amg was used and the results show that hex solution does not scale that well. The reasons are still need to be discovered.