Traffic prediction is a fundamental and vital task in IntelligenceTransportation System (ITS), but it is very challenging to get high accuracywhile containing low computational complexity due to the spatiotemporalcharacteristics of traffic flow, especially under the metropolitancircumstances. In this work, a new topological framework, called LinkageNetwork, is proposed to model the road networks and present the propagationpatterns of traffic flow. Based on the Linkage Network model, a novel onlinepredictor, named Graph Recurrent Neural Network (GRNN), is designed to learnthe propagation patterns in the graph. It could simultaneously predict trafficflow for all road segments based on the information gathered from the wholegraph, which thus reduces the computational complexity significantly from O(nm)to O(n+m), while keeping the high accuracy. Moreover, it can also predict thevariations of traffic trends. Experiments based on real-world data demonstratethat the proposed method outperforms the existing prediction methods.