It is definitely an interesting problem how to sample uniformly from the set of n-dimensional vectors with an L2-norm smaller or equal to 1. I am sure that it has practical applications, too. Also, nobody said that n has to be huge.