On the on-line maintenance scheduling problem

machine instantly serves requests but needs to undergo maintenance after serving a maximum of L requests. We want to maximize the number of requests served. In the on-line version, we prove that serving L requests before placing a maintenance is 0.5-competitive and is best possible for deterministic algorithms. We describe a 0.585-competitive randomized algorithm and show an upper bound of 2L/(3L-1) . We also analyze the empirical performance of various on-line algorithms on specific arrival distributions.