Wednesday March 08, 2006

Sun has decided to join forces with the Open MPI community by becoming a core, contributing community member. As such, we will contribute to the Open MPI code base and use it to deliver future supported versions of MPI for Solaris. This is big news!

Sun's history with MPI really began in late 1996 when a group of us came to Sun from Thinking Machines where we'd been working on a multi-platform parallel programming toolkit called GlobalWorks. GlobalWorks was based in large part on technology we had developed for TMC's Connection Machine massively parallel supercomputers through the 1980s into the mid-1990s.

Over the intervening years at Sun, we put a huge effort into our MPI library. Early on we abandoned our original MPICH base and did a clean sheet design. Our implementation was one of the first to support the full MPI2 standard, including a nicely designed, fine-grain approach to thread safety and a robust client-server capability In addition, because our HPC efforts at the time were centered around clusters of very fat SMP nodes (up to 100 CPUs per node), we put significant effort into our shared memory optimizations for fast data transfer and collectives.

By joining the Open MPI community we will bring our expertise, experience, and resources to that effort while also benefiting from the work of other members of the community. Los Alamos National Laboratory (LANL), Indiana University (IU), the University of Tennessee (UTK), the University of Houston, and the University of Stuttgart (HLRS) all bring excellent MPI experience and expertise to the community. In addition, with Cisco Systems also joining Open MPI there will now be two vendors bringing an intense, product-oriented focus to the effort which should be of great benefit to the community and its user base. Including our customers.

This shift towards open source is directly aligned with our belief that because the creation of ultra-scale software tools components for HPC is so expensive, such development should be done in the commons rather than in separate, duplicative efforts. It also aligns with our belief that innovation does happen elsewhere and we should leverage it in cases where that makes sense, as it does here.