Mobile Internet traffic keeps a fasting-growth globally. This paper is to shed light on the traffic analysis and policy control in EV-DO systems. Large-scale datasets collected from an urban area are studied for characterizing the application/subscriber usage and downlink-uplink imbalance. Our observations show that HTTP and P2P contribute over 80% traffic, due to the prevalence of video applications. We also noticed that the traffic volume distribution of subscriber fits the Zipf law with Zipf (.38) and Zipf (1.06), and the downlink-uplink ratio of the whole dataset is 4.37, while HTTP and P2P are 11.8 and 2.41 respectively. With these numerical analysis results, we formulate a traffic control scenario, and present its optimization solution. Finally, several policy control considerations are discussed from the perspective of network operator.