Understanding the Behavior and Performance of Non-blocking Communications in MPI

The behavior and performance of MPI non-blocking message passing operations are sensitive to implementation specifics as they are heavily dependant on available system level buffers. In this paper we investigate the behavior of non-blocking communication primitives provided by popular MPI implementations and propose strategies for these primitives than can reduce processor synchronization overheads. We also demonstrate the improvements in the performance of a parallel Structured Adaptive Mesh Refinement (SAMR) application using these strategies.

• Type: Conference
• Year: 2004
• Where: EUROPAR
• Authors: Taher Saif, Manish Parashar
• Keywords: EUROPAR 2004, MPI Non-blocking Message, Non-blocking Communication Primitives, Popular Mpi Implementations