Tutorial Details: Parallel Programming Using MPI
|Date:||Tuesday, March 3, 2009, 10:00 am - 04:00 pm|
|Instructor(s):||David Porter, MSI, Hakizumwami Birali Runesha, MSI|
This one-day workshop on MPI will help researchers write better and portable parallel codes for distributed-memory machines like Linux clusters, such as MSI's 248-core Intel-based cluster (Calhoun), the 1,000-core AMD cluster (Blade), and the SGI Altix Itanium-based shared-memory system. It will focus on basic point-to-point communication and collective communications, which are the most commonly used MPI routines in high-performance scientific computation. In addition, the advantage of using MPI non-blocking communication will be introduced. Each session of the workshop will combine a lecture with hands-on practice. The lecture will introduce basic principles, and the hands-on portion will focus on the use of MPI principles via examples.
Session One: Introduction to basic concepts of "MPI is Small," centering on point-to-point communication.
Session Two: MPI collective communications including broadcast, gather, scatter, and Alltoall. Programming will be done in Fortran and C, so any background in these two languages will be helpful.
|Prerequisites:||Familiarity with UNIX/Linux and knowledge of Fortran, C, or C++|