MPI is the de-facto standard for message passing in parallel scientific applications. MPI-IO is a part of the MPI-2 specification defining file I/O operations in the MPI world. MPI-IO enables performance optimizations for collective file I/O operations as it acts as a portability layer between the application and the file system. The goal of this study is to optimize collective file I/O operations. Three different algorithms for performing collective I/O operations have been developed, implemented, and evaluated on a PVFS2 file system and over NFS. The results indicate that different algorithms promise the highest write bandwidth for different number of processors, application settings and file systems, making a one-size-fits-all solution inefficient.

No abstract