2019 · 50 citations · 20 references
ComputingComputer ScienceComputer SystemsComputer SecurityCryptographyNetwork SecurityIntrusion DetectionLatency OptimizationPrivacy-preserving ProtocolsDeep Packet InspectionEncrypted Traffic AnalysisEncrypted Traffic InspectionPacket InspectionRule-based GenerationRule-based Pattern GenerationSetup Delay
Network middleboxes perform deep packet inspection (DPI) to detect anomalies and suspicious activities in network traffic. However, increasingly these traffic are encrypted and middleboxes can no longer make sense of them. A recent proposal by Sherry et al. (SIGCOMM 2015), named BlindBox, enables the middlebox to perform inspection in a privacy-preserving manner. BlindBox deploys garbled circuit to generate encrypted rules for the purpose of inspecting the encrypted traffic directly. However, the setup latency (which could be 97s on a ruleset of 3,000 as reported) and overhead size incurred by garbled circuit are high. Since communication can only be commenced after the encrypted rules being generated, such delay is intolerable in many real-time applications. In this work, we present PrivDPI, which reduces the setup delay while retaining similar privacy guarantee. Compared to BlindBox, for a ruleset of 3,000, our encrypted rule generation is 288x faster and requires 290,227x smaller overhead for the first session, and is even 1,036x faster and requires 3424,505x smaller overhead over 20 consecutive sessions. The performance gain is based on a new technique for generating encrypted rules as well as the idea of reusing intermediate results generated in previous sessions across subsequent sessions. This is in contrast to Blindbox which performs encrypted rule generation from scratch for every session. Nevertheless, PrivDPI is 6x slower in generating the encrypted traffic tokens, yet in our implementation, the token encryption rate of PrivDPI is more than 17,271 per second which is sufficient for many real-time applications. Moreover, the intermediate values generated in each session can be reused across subsequent sessions for repeated tokens, which could further speedup token encryption. Overall, our experiment shows that PrivDPI is practical and especially suitable for connections with short flows.
20
Practical techniques for searches on encrypted data
Dawn Xiaoding Song, David Wagner, Adrian Perrig · 2002 · 3.3K citations
Security and Composition of Multiparty Cryptographic Protocols
Ran Canetti · Journal of Cryptology · 2000 · 1.4K citations · Full text
Efficient Garbling from a Fixed-Key Blockcipher
Mihir Bellare, Viet Tung Hoang, Sriram Keelveedhi et al. · 2013 · 307 citations · Full text
Justine Sherry, Chang Lan, Raluca Ada Popa et al. · 2015 · 285 citations · Full text
Machine Learning for Encrypted Malware Traffic Classification
Blake Anderson, David McGrew · 2017 · 269 citations