Abstract: Internet is a complex network with the characteristics of self-organized criticality. If the nodes with many more connections are attacked, it may lead to dropped efficiency and even abnormal due to overload. The successive traffic on the overloaded nodes is compelled to reroute to avoid the congested nodes. The bypassing may congest other downstream nodes and lead to more traffic detour and node congestion, and then a cascading failure may happen. The cascading dynamics of the Internet are analyzed and two reasons are pointed out which may cause cascading failures. Different from betweenness centrality, the congestion function to represent the congested extent of node is proposed. By introducing the concept of “delay time”, the correlation between permanent removal and non-removal is built, and the flexibility of model is improved. And a new assessing function of network efficiency based on congestion effects is given in order to measure the destruction of cascading failures, which highlights a more meaningful way to measure the damage of cascading failures. Moreover, some impacts of structure and size of topology, delay time, handling capability and generating speed of packets on congestion propagation are also investigated, and congestion propagation process consisting of three phases and some factors affecting transition phenomenon are uncovered.