Title: A New Flexible MPI Collective I/O Implementation

Speaker: Kenin Coloma, NECIS Student, Northwestern University

Location: CSRI Room 90 (Sandia NM)

Brief Abstract: MPI is undoubtedly a critical component of any cluster computing system without which the power of these systems would be much less accessible. As the number of processors in clusters and more specialized parallel systems grows, I/O soon becomes a bottleneck for many applications ranging from climate modeling to computational physics. The MPI-2 collective I/O routines were included for the sole purpose of optimization. Optimizations for this set of I/O functions can easily become very complex behind the scenes, sometimes limiting both the possible parameter variations and the number of developers and contributors. Given the pervasiveness and source code availability of the ROMIO implementation, our new implementation seeks to provide the equivalent behavior with respect to the two-phase I/O optimization while simultaneously delivering a cleaner code base and more avenues for research. The key goals for the new implementation are flexibility, performance, and developer friendliness. These compose an ideal research platform and important framework for further exploration. As an example, several variations on the two-phase I/O optimization are explored.

CSRI POC: Lee Ward, (505) 844-9545