Hybrid MPI (HMPI) is an MPI implementation that extends key MPI design details and interface restrictions to achieve high performance on multi- and many-core CPUs via novel shared memory techniques. Higher core counts per compute node mean more and more execution contexts are connected by shared memory rather than a network, making shared memory communication for parallel computing an important area of research. HMPI's layered library design provides an excellent tool for application developers as well as a platform for future research and development on high-performance MPI-based communication in heterogeneous systems. Research involves new algorithmic techniques that 1) improve communication performance in shared memory environments and 2) expand the scope of HMPI to other devices such as Xeon Phi accelerators and graphical processing units (GPUs).
The distribution of open source software is one way that CREST contributes to the wealth of scientific research at Indiana University and throughout the country. The production of open source software significantly enhances the value of scholarly work due to reproducibility andability of scientific research. The open source software created by CREST and its affiliates boosts IU’s competitive footing for federal grant funding, which in turn, brings federal research monies and new high-quality jobs into the state. Also used in the private sector, these open source software products aid in the development of the Indiana economy.