DDoS attack has caused severe damage to e-commerce and e-service and is great intimidation to the development of them. After analyzing the detection algorithm of D-WARD, a representative source-end DDoS detection system, we introduced a nonparametric change point detection method in statistics and improved D-WARD with nonparametric recursive CUSUM algorithm and compared their respective performance. Experiments proved that the improved system is lower in false-positive rate and false-negative rate, which is more accurate and could adapt to more complex network environments. This improved system could be utilized in the protection of e-commerce against DDoS attacks.