Traffic classification methods developed rapidly. Most of these methods have focused on high flow accuracy. However, the majority of the total bytes are occupied by a small part of flows. This flows may include P2P streaming, P2P file sharing, worms or DoS attack etc. Their effect to the internet is far more than the general applications. In this paper, we first analyze this type of applications and obtain significantly differences between them. And based on the analysis results we develop a classifier. Moreover, we design two methods to accelerate the classification performance, TCP packet accelerate and UDP packet heuristics method. Our experiment result shows that this classifier can classify the majority of total traffic.