In this paper we study packet transmission strategies for data service over wireless networks. We assume the wireless channel is in either a good or a bad state and that transferring a packet under a good channel state consumes less power than under a bad channel state. Under the Markov channel assumption, it is proved that the optimal strategy is of threshold structure: Corresponding to each channel state there exists a positive threshold on queue length; once the queue length exceeds the threshold, the network should transmit packets over its capacity; otherwise there should be no transmission. A single sample path-based optimization algorithm is proposed to tune the thresholds. Since only a single sample path is involved, the proposed algorithm can be implemented online.