Comparison

How does the performance between shared memory parallel computing and MPI applications compare?

MPI is faster. This is because there is less of an overhead than with libraries like openMP. However, MPI requires more modifications to code, and thus is usually harder to implement. This difference in learning curve is usually why shared memory parallel computing is still widely implemented rather than solely using MPI. There is another reason why MPI is not widely used over shared memory parallelism, which is explained on the next page.