Simulations, for example to determine the distribution of the mean time between failure (see the simple MATLAB/Octave script mtbf.m), as done by Monte Carlo methods are another kind of embarrassingly parallel computations. We sketched two potential problems: (1) we must ensure that every processor generates different random sequences which are independent of what is used on other computers; and (2) for large number of simulations, the periodicity of the random number generator may prevent the increase of the accuracy of the results.
Bibliography