In this paper we investigate a model for delay-optimal scheduling for networks using relay station (RS) in contemporary broadband wireless access networks. We present an optimal control problem using a queueing model based on Markov Decision Process (MDP) incorporating an opportunistic scheduling for downlink in a RS architecture. Using a simplified queueing model for access networks, we propose a policy for the case where the arrival and channel statistics are symmetric among users. The key idea behind the policy is that users should greedily exploit opportunism at RS, while the basestation should prioritize on balancing queues at RS. An analogous fluid problem is presented to show that such policy is optimal in the fluid regime. Through numerical methods we discuss the performance of the proposed scheme and another throughput-optimal solution such as backpressure policies.