This paper presents a distributed system for real-time anomaly detection in backbone. The backbone traffic is so huge that it is difficult to monitor abnormal traffic by traditional methods. Our system is based on Hadoop, an open source framework, and used to detect the abnormal traffic. Firstly, We establish a precise regression model to describe the network traffic. Secondly, distributed system Hadoop is used to detect the abnormal traffic. Finally, the experimental results prove that our system can detect the abnormal traffic accurately and efficiently in the real-world network environment.