Modern HEC systems, such as Blue Gene/P, rely on achieving high-performance by using
the parallelism of a massive number of low-frequency/low-power processing cores. This
means that the local pre-and post-communication processing required by the MPI stack
might not be very fast, owing to the slow processing cores. Similarly, small amounts of
serialization within the MPI stack that were acceptable on small/medium systems can be
brutal on massively parallel systems. In this paper, we study different non-data …

Each MPI sent message carries a tag, and each received message carries a tag and
information about the source. So the receiver needs to search the queue of posted receive
requests to find the one that matches the arrived message. Most current MPI
implementations use a single queue for receive requests. This has a potential scalability
problem when the length of queue becomes large.