Practical Analysis of Encrypted Network Traffic Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 19, 2019
  • White, Andrew
    • Affiliation: College of Arts and Sciences, Department of Computer Science
  • The growing use of encryption in network communications is an undoubted boon for user privacy. However, the limitations of real-world encryption schemes are still not well understood, and new side-channel attacks against encrypted communications are disclosed every year. Furthermore, encrypted network communications, by preventing inspection of packet contents, represent a significant challenge from a network security perspective: our existing infrastructure relies on such inspection for threat detection. Both problems are exacerbated by the increasing prevalence of encrypted traffic: recent estimates suggest that 65% or more of downstream Internet traffic will be encrypted by the end of 2016. This work addresses these problems by expanding our understanding of the properties and characteristics of encrypted network traffic and exploring new, specialized techniques for the handling of encrypted traffic by network monitoring systems. We first demonstrate that opaque traffic, of which encrypted traffic is a subset, can be identified in real-time and how this ability can be leveraged to improve the capabilities of existing IDS systems. To do so, we evaluate and compare multiple methods for rapid identification of opaque packets, ultimately pinpointing a simple hypothesis test (which can be implemented on an FPGA) as an efficient and effective detector of such traffic. In our experiments, using this technique to “winnow”, or filter, opaque packets from the traffic load presented to an IDS system significantly increased the throughput of the system, allowing the identification of many more potential threats than the same system without winnowing. Second, we show that side channels in encrypted VoIP traffic enable the reconstruction of approximate transcripts of conversations. Our approach leverages techniques from linguistics, machine learning, natural language processing, and machine translation to accomplish this task despite the limited information leaked by such side channels. Our ability to do so underscores both the potential threat to user privacy which such side channels represent and the degree to which this threat has been underestimated. Finally, we propose and demonstrate the effectiveness of a new paradigm for identifying HTTP resources retrieved over encrypted connections. Our experiments demonstrate how the predominant paradigm from prior work fails to accurately represent real-world situations and how our proposed approach offers significant advantages, including the ability to infer partial information, in comparison. We believe these results represent both an enhanced threat to user privacy and an opportunity for network monitors and analysts to improve their own capabilities with respect to encrypted traffic.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Reiter, Michael
  • Bailey, Michael
  • Jeffay, Kevin
  • Porras, Phillip
  • Monrose, Fabian
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2015
Place of publication
  • Chapel Hill, NC
  • There are no restrictions to this item.

This work has no parents.