SGI MPI is most usable on a single host, such as a large Origin 2000, and is slightly less usable in clusters. Because of the array services software, MPI applications are managed as a single unit, even when spread across multiple nodes. Starting an application on a single host is easy, but the syntax for starting on multiple hosts is somewhat painful, and isn't managed directly by the array services software.
Standard I/O is fine when using a single host but is currently not handled well when using multiple hosts. Output from processes on remote nodes is lost and there is no optional labeling of output lines by process number. These will be addressed in a forthcoming release.