Simple Parallel Data Structures

• Getting started with "Hello World"
• Sharing Data
  • Using MPI datatypes to share data
  • Using MPI_Pack to share data
• Sending in a ring (broadcast by ring)
  • Using topologies to find neighbors
• Finding PI using MPI collective operations
• Fairness in message passing
  • Implementing Fairness using Waitsome
• A Parallel Data Structure
  • Using nonblocking operations
  • Shifting data around
  • Exchanging data with MPI_Sendrecv
• A simple Jacobi iteration
• Collecting Data
  • Collecting Data (with varying amounts from each process)
• Putting it all together: A complete application

Master/slave

• Master/slave programs in MPI
• A simple output server

Performance tuning MPI

• Determining delivered memory performance
  • Determining delivered memory performance with unaligned data
• Benchmarking point to point performance
  • Benchmarking point to point performance with MPI_Ssend
  • Benchmarking point to point performance with nonblocking operations
  • Benchmarking point to point performance with nonblocking operations, head-to-head
  • Benchmarking point to point performance with unaligned data
  • Benchmarking point to point performance with contention
  • Benchmarking point to point performance with contention with MPI_Ssend
• Determining the amount of MPI buffering
• Benchmarking collective barrier
  • Benchmarking collective Allreduce
• Comparing the performance of MPI vector datatypes
• Exploring the cost of synchronization delays
• Pipelining pitfalls
• Demonstration of poor performance caused by unnecessary synchronization
  • Demonstation of improved performance by defering synchronization
• Jacobi Iteration - Example Parallel Mesh
  • Jacobi Iteration - Shift up and down
  • Jacobi Iteration - Exchange head-to-head
  • Jacobi Iteration - Nonblocking send/recv
  • Jacobi Iteration - Nonblocking send/recv for receiver pull
  • Jacobi Iteration - Synchronous send
  • Jacobi Iteration - Ready send
  • Jacobi Iteration - Nonblocking Ready send
  • Jacobi Iteration - Overlapping communication
  • Jacobi Iteration - Overlapping communication (sends first)
  • Jacobi Iteration - Persistent send/recv

Miscellaneous Advanced MPI

• Using Datatypes to Scatter Matrices Among Processors, also in Fortran