04_pure-mpi/02_mpingpong

Hands-on: IMB: Multiple Pingpong (Send and Recv)

• Both of C/C++ and Fortran users will perform the common benchmarks, although IMB is written by C and C++.

How to execute

1. Edit a job script

• Before trying this hands-on, you need to do 00_imb to compile Intel MPI benchmark.
• We have two working directories, cs32cl256_P48Q2 and cs32cl256_P4Q2.
• A job script to execute the program is task.sh in each of working directories.
• Edit BINDIR variable in task.sh. You need to write your installed location of IMB binary (e.g., IMB-MPI1) there.

2. Run program

• You can run the program either:

## To run as a batch job
$ cd cs32cl256_P48Q2
$ pjsub task.sh 
## Or, to run in an interactive job
$ cd cs32cl256_P48Q2
$ bash task.sh 

• The jobs in the Exercises will be completed within 1-2 minutes.
  • For safety, we set the elapsed time of the job scripts as 4 minutes.

Exercises A

• E1: Check the node and MPI settings in cs32cl256_P48Q2/task.sh and cs32cl256_P4Q2/task.sh. What kinds of differences are there?
  • P48Q2: 2 nodes, 48 processes per node.
  • P4Q2 : 2 nodes, 4 processes per node. Four is equivalent to the number of CMG per node.
• E2: Compare the measured bandwidth between the two settings (i.e., P48Q2 and P4Q2).

Exercises B (advanced)

• E3: Confirm whether MPI communication between non-neighboring nodes occurs or not, using the MPI statistical information.