The challenges of providing data services over wireless networks come from, their high reliability requirement under noisy wireless channels. In this paper, we model the noisy channel by a continuous-time Markov process and propose a dynamical packet transmission strategy for data service. The transmission strategy optimization is formulated as a Markov decision process. We establish that the optimal strategy has a threshold structure and develop a single sample path-based optimization algorithm to tune the thresholds. Since only a single sample path is involved, the proposed algorithm can be implemented online.