AASC: Anonymizing Network Addresses based on Subnet Clustering

The network packet trace dataset plays an important role in networking research. Publishing those traces publicly faces how to protect the providers' sensitive privacy, especially the internal IP addresses. In this paper, we propose a subnet-clustering based method, AASC, to anonymize those internal addresses. According to AASC, three parts of a whole IP address are anonymized by different methods. The network part is anonymized by a prefix-preserved anonymization method, the subnet part is generalized by clustering based on a predefined set of port numbers, and the host address is randomized. We also define two entropy based metrics, the simple measure and the co-existence measure, to measure the degree of privacy preserved in anonymized addresses. The defined metrics can reflect some dependencies among trace records. We develop a local-search based, measure-guided algorithm to search subnet clusters with more utilities. We have conducted some experiments to validate our proposed method.