You don't need to be an overlay journal to be an open access journal. In machine learning (that maybe is something many here are more interested in than the quite abstract math stuff of Discrete Analysis), JMLR has been around for a quite long time: http://jmlr.csail.mit.edu/papers/