random_mpi, a C++ code which demonstrates one way to generate the same sequence of random numbers for both sequential execution and parallel execution under MPI.
A simpler approach to random numbers would simply let each processor choose a seed. Or the master processor could choose distinct seeds. However, this is not ideal since it will not match the sequential program and it does not avoid the possibility that two of the random sequences will quickly overlap because of a bad choice of seed.
Notice that if we have 10 processors available under MPI, we do not want each processor to generate the same random number sequence. Instead, we want each of the processors to generate a part of the sequence, so that all the parts together make up the same set of values that a sequential program would have computed.
We assume we are using a linear congruential random number generator or "LCRG", which takes an integer input and returns a new integer output:
U = ( A * V + B ) mod CWe assume that we want the MPI program to produce the same sequence of random values as a sequential program would - but we want each processor to compute one part of that sequence.
We do this by computing a new LCRG which can compute every P'th entry of the original one.
Our LCRG works with integers, but it is easy to turn each integer into a real number between [0,1].
The particular scheme for computing the parameters of the new LCRG is implemented in the UNIFORM library.
The latest versions of MPI no longer support the special C++ MPI bindings, so the examples given here have reverted to using the C MPI bindings.
The information on this web page is distributed under the MIT license.
random_mpi is available in a C version and a C++ version and a Fortran90 version.
mpi_test, a C++ code which uses the message passing interface (MPI) for parallel computations in a distributed memory environment.