One of the most challenging issues in cognitive radio networks is efficient channel sensing and channel accessing. In this paper, an analytical queueing model is used to derive the probability of successful transmission, channel sensing time, and transmission quota, for each data channel. Each CR node records the derived statistics in a channel preference matrix. A CR pair selects a data channel for sensing and accessing based on the successful transmission probability. According to the derivations, we design a media access control protocol, which utilizes the powerful computation capability of cloud servers to estimate the behavior of PUs, for infrastructure-based cognitive radio networks. We validate the analytical model with simulation results. Besides, the proposed MAC protocol is compared with other approaches via simulation. The simulation results showed that our protocol performs well in both utilization of channel idle time and the average tries of channel search.