An Empirical Study of an Anonymity Metric for Data Networks




Vasudevan, Abinash

Journal Title

Journal ISSN

Volume Title



Privacy and data protection are two very important needs in the modern day Internet. One of the attacks to privacy is eavesdropping, i.e., an outsider or an attacker listens to a private conversation and identifies the people involved in the conversation. In order to protect the content which is to be transmitted, encryption methods are used. Even if the data is encrypted, it is possible for the attacker to identify the end user, i.e., the person sending the data and the person to whom the data is sent. An anonymous network is a type of network that prevents traffic analysis and protects the identity of the end users. Some of the popular anonymous networks, namely, TOR, etc., are used for identity concealment. These anonymous networks provide real-time, low latency anonymous communications. Because of the low latency implementation, timing constraints are imposed on the low-latency networks. Due to this timing constraint, an attacker will be able to get details from the packet timing information and use it to identify the end users. This makes the timing attack possible in these low-latency anonymous networks. In this thesis, an anonymity metric that can measure the practical effectiveness of low-latency anonymous network is studied. The anonymity metric calculates the timing distortion between the incoming packets and the outgoing packets and uses wavelet-based Multi-Resolution Analysis (MRA) to determine the anonymity of the network. For the purpose of analysis, packet traffic is simulated based on several stochastic traffic models. The simulated traffic contains information about the timing of packets entering and leaving the anonymous network. The most basic traffic model used for this purpose is the Poisson process; we also use the Markov Modulated Poisson Process and the Markovain Arrival Process for further analysis. The end-to-end network delay is characterized with various probability distributions for this study. The results of the analysis show that the measured energy used to compute the anonymity metric is higher for more complex traffic arrival models, which implies that the anonymity level is correspondingly higher. The more complex traffic patterns introduce more randomness in the timing information, resulting in higher measured energy values. Hence, for example, the Poisson process arrival model has the least energy among the traffic models used in this study.



Anonymous networks, Privacy, Anonymity metric, Anonymity