Abstract
Introducing over-decomposition at the application level to pre-existing codes can be challenging. We use Adaptive MPI, an implementation of MPI on top of Charm++ that provides application-independent support for processor virtualization, dynamic load balancing, and fault tolerance. We port PlasComCM, the main simulation code of the PSAAPII center at UIUC (XPACC), written in Fortran90 and MPI, to AMPI and show optimization and performance studies enabled by AMPI.