[Trilinos-Users] Zoltan question

[Biddiscombe, John A.]

Erik

>Zoltan assumes "everyone" has MPI on parallel machines so there is no
>abstraction layer above that. However, Zoltan only uses a small subset
>of MPI so you could implement that subset in your own messaging library
>and link with that as a "fake" MPI library. We use that approach for
>serial builds, so it should work.

OK. That sounds quite doable. I’ll have a go at that.

Would that same argument hold for zoltan2, or are the dependencies on
other parts of trilinos such that it would become too complicated as they
all pull in other bits of MPI?

Ta

JB