A novel attack detection scheme is proposed to prevent web proxy based http attacks.Here the detection is carried out at server.The scheme utilize locality behaviors such as temporal and spatial localities to obtain features of proxy-to-server traffic. A Gaussian-mixture and Gamma distribution hidden semi markov model is used to describe the time-varying traffic behavior of web proxies. Soft control scheme is used as the attack response it convert malicious traffic into relatively normal one by behavior reshaping rather than rudely discarding so that legitimate users get service and prevent attacker from receiving service.DDoS attack is a key threat to internet applications by comparing existing network security systems the primary aim of the proposed system is to protect origin server from web proxy based http attacks.