Distributed attacks have reportedly caused the most serious losses in the modern cyber environment. Thus, how to avoid and detect distributed attacks has become one of the most important topics in the cyber security community. Of many approaches for avoiding and detecting cyber-attacks, behavior based method has been attracting great attentions from many researchers and developers. It is well known that, for behavior based cyber-attack detections, the algorithm for extracting normal modes from historic traffic is critically important. In this paper, after the newest algorithms for extracting normal behavior mode from historic traffics are discussed, a novel algorithm is proposed. Its efficiency is examined by experiments using darknet traffic data.