Largest unstructured mesh problem solved till date

What is the largest problem (in DOF) that has been solved till date using a fully unstructured mesh. I know about PFLOTRAN but I am not sure if it is the largest. If someone can point to relevant papers then that would be great.

Actually I need some pointers on creating and reading in large meshes on 4K plus cores. PFLOTRAN papers have some ideas but I am wondering if there is anything else.

1 Answer

We have solved problems with 2 billion unknowns in deal.II, so this is a lower bound. The problem wasn't the size but that we ran out of signed integers. I've heard anecdotally that others have solved problems with 100 billion unknowns.

I don't think that any of these computations happen on completely unstructured meshes. Instead, a coarse mesh with maybe a few 100,000 or millions of cells is generated and then refined in some structured way to obtain the final mesh.

• To follow up on my own reply, I've had a conversation with Ulrich Ruede the other day who showed me results of computations in which they've solved problems with $10^{12}$ unknowns. Jun 6 '12 at 21:17