Information fusion-based method for distributed domain name system cache poisoning attack detection and identification
In this study, the authors consider the detection and identification problems of distributed domain name system (DNS) cache poisoning attack. In the considered distributed attack, multiple cache servers are invaded simultaneously and the attack intensity for each cache server is slight. It is difficult to detect and identify the distributed attack by the existing local information-based detection methods, as the abnormal features for each cache server are indistinctive under distributed attack. To handle this problem, they propose an information fusion-based detection and identification methods. They find that the entropies of the query Internet protocol (IP) addresses for all cache servers are approximately stationary and statistically independent under normal cases. When distributed attack happens, they show the fact that the correlation of the entropies among all cache servers could increase dramatically. On the basis of this feature, they make use of principal component analysis to design the detection and identification methods. Specifically, attack is true when the maximum eigenvalue of the normalised entropies matrix exceeds a threshold, and the attacked servers are identified by the main loading vector. At last, they take a large-scale DNS in China and a simulation as two examples to show the effectiveness of their methods.