On the optimality of Least Recently Used

It is well known that competitive analysis yields results that do not reflect the observed performance of online paging algorithms. Many deterministic paging algorithms achieve the same competitive ratio, ranging from inefficient strategies as flush-when-full to the well-performing leastrecently-used (LRU). In this paper, we study this fundamental online problem from the viewpoint of stochastic dominance. We give simple proofs that when sequences are drawn from distributions modelling locality of reference, LRU stochastically dominates any other online paging algorithm. As a byproduct, we obtain simple proofs of some earlier results.