Lecture 24

Learning objectives

After this class, you should be able to:

1. Write simple MPI code and run it on multiple processes on XSEDE machines.

Reading assignment

1. Slides on Running MPI.
2. UIUC Lecture 26.
3. MPI tutorial at: https://computing.llnl.gov/tutorials/mpi

Exercises and review questions

• Exercises and review questions on current lecture's material
  1. Write MPI code so that processes work together to add numbers of a large array. Run this code on Keeneland. Report time as a function of the number of processes on the discussion forum.

• Preparation for the next lecture
  1. Modify the above code so that each process performs its additions on the GPU before the final sum is computed on the CPU. Compare its performance against that of the above code. If you don't report the results of this task, then you will need to combine CUDA and MPI in your project.

---

Last modified: 9 Apr 2013