Dear all,

 

When adding #include "mpi.h" and building a single process with mpi, I find that the single process mpi job runs very much slower than a single thread process built with gcc. i.e. 150 to 200% slower, and run as a standalone executable.

 

That is, for example

 

gcc mycode.c -o mycode.exe

./mycode

 

 

will run in 100 seconds.

 

 

Adding #include "mpi.h" to mycode.c

 

mpicc mycode.c -o mycode.exe

mpirun -n 1 mycode.exe

 

 

will take 180 to 400 seconds to run.

 

This varies from system to system on intel, Sparc or IBM based systems.

 

Can anyone explain this offhand ? Has anyone else experienced this ?