The increase of encrypted traffic on the Internet may become a problem for network-security applications such as intrusion-detection systems or interfere with forensic investigations. This fact has increased the awareness for traffic analysis, i.e., inferring information from communication patterns instead of its content. Deciding correctly that a known network flow is either the same or part of an observed one can be extremely useful for several network-security applications such as intrusion detection and tracing anonymous connections. In many cases, the flows of interest are relayed through many nodes that reencrypt the flow, making traffic analysis the only possible solution. There exist two well-known techniques to solve this problem: passive traffic analysis and flow watermarking. The former is undetectable but in general has a much worse performance than watermarking, whereas the latter can be detected and modified in such a way that the watermark is destroyed. In the first part of this dissertation we design techniques where the traffic analyst (TA) is one end of an anonymous communication and wants to deanonymize the other host, under this premise that the arrival time of the TA's packets/requests can be predicted with high confidence. This, together with the use of an optimal detector, based on Neyman-Pearson lemma, allow the TA deanonymize the other host with high confidence even with short flows. We start by studying the forensic problem of leaving identifiable traces on the log of a Tor's hidden service, in this case the used predictor comes in the HTTP header. Afterwards, we propose two different methods for locating Tor hidden services, the first one is based on the arrival time of the request cell and the second one uses the number of cells in certain time intervals. In both of these methods, the predictor is based on the round-trip time and in some cases in the position inside its burst, hence this method does not need the TA to have access to the decrypted flow. The second part of this dissertation deals with scenarios where an accurate predictor is not feasible for the TA. This traffic analysis technique is based on correlating the inter-packet delays (IPDs) using a Neyman-Pearson detector. Our method can be used as a passive analysis or as a watermarking technique. This algorithm is first made robust against adversary models that add chaff traffic, split the flows or add random delays. Afterwards, we study this scenario from a game-theoretic point of view, analyzing two different games: the first deals with the identification of independent flows, while the second one decides whether a flow has been watermarked/fingerprinted or not.
Traffic Analysis, Network Forensics, Intrusion Detection, Anonymous Networks
Prince of Asturias Chair of the University of New Mexico, sponsored by the Iberdrola Foundation.
Level of Degree
Electrical and Computer Engineering
First Committee Member (Chair)
Second Committee Member
Third Committee Member
Fourth Committee Member
Elices Crespo, Juan Antonio. "Neyman-Pearson Decision in Traffic Analysis." (2014). https://digitalrepository.unm.edu/ece_etds/76