High efficiency cross-layer resource management is desired for real-time multimedia streaming application in cognitive radio network. In this paper, a novel multi-user resource management scheme is proposed. Its working principle and information swap mechanism are illustrated in detail. The proposed RDMAL (real time distributed multi-agent learning) algorithm takes advantage of multi-agent learning approach and relies on it to dynamically exploit available frequency channel, while uses available interference information to achieve the learning efficiency. The simulation and preliminary experimental results demonstrate that the proposed RDMAL can achieve lower packet loss rate than state-of-the-art dynamic resource management algorithm and it can decrease the time overheads in terms of the cost of information exchange.