140 likes | 245 Views
Enabling Information Confidentiality in Publish/Subscribe Overlay Services. Hui Zhang Haifeng Chen Guofei Jiang Xiaoqiao Meng Kenji Yoshihira NEC Labs America Abhishek Sharma University of Southern California. Outline. Problem statement
E N D
Enabling Information Confidentiality inPublish/Subscribe Overlay Services Hui Zhang Haifeng Chen Guofei Jiang Xiaoqiao Meng Kenji Yoshihira NEC Labs America Abhishek Sharma University of Southern California
Outline • Problem statement • Information confidentiality in pub/sub overlay services • Information foiling • Mechanism description • Performance metrics • Fake message generation schemes • Evaluation • Conclusions & future work
Publish/Subscribe overlay services Subscription Publisher X Event Subscriber A Broker network Subscriber B Publisher Y
Information confidentiality in pub/sub services • Publish/subscribe decouples publishers and subscribes. • Events are characterized into classes, without knowledge of what (if any) subscribers there may be. • Subscribers express interest in one or more classes, and only receive messages that are of interest, without knowledge of what (if any) publishers there are. • New confidentiality problems in this content-based routing process • Can the broker network perform content-based routing without the publishers trusting the broker network with the event content? • Information confidentiality • Can subscribers obtain dynamic data without revealing their subscription functions (content) to the publishers or broker network? • Subscription confidentiality • Can publishers control which subscribers may receive particular events? • Publication confidentiality
Problem definition • Formulation of pub/sub confidentiality as a communication problem. • Upon an event e, the broker determines if each subscription s in the active subscription set matches the event based on a function f(e; s), but without learning the information contained in e and s. • Threat model: a broker is assumed to be computationally bounded and exhibits a semi-honest behavior.
Information foiling – the mechanism • Subscriber: for each active subscription, generates ks foiling subscriptions, and send them in a random order to the broker which store them all as active subscriptions. • Publisher: for each event, generates kp foiling events, and send them in a random order to the broker. • Broker: upon each arriving event e, decides the subset of the active subscription set and send one notification for each matched subscription. • (optional) Subscriber: upon a notification associated with one authentic subscription, sends a confirmation request to the publisher. • (optional) Publisher: upon a confirmation request, sends a reply to the subscriber upon the authenticity of the related event.
Information foiling – performance metrics • Assume the attacker has a function F : f{e, Ee} -> G, that takes the composite message set {e, Ee} as input and outputs a message set G {e, Ee} consisting of messages that the attacker perceives as useful. • Metric 1: indistinguishability • defined as , where I(e, G) = 1 if e 2 G; 0 otherwise. • Metric 2: truth deviation • dened as , where D(e, g) is the difference between the values of messages e and g. • Metric 3: communication overhead • it depends not only on the information foiling mechanism but also on the actual data distributions of the authentic events and subscriptions.
Fake message generation – a probabilistic model • Consider an event message m with L attributes. • Let the value Vi for attribute Ai in m be a random variable taking values in V according to a probability mass function pVi . • Let Vm = (V1, V2, …, VL), represent m, i.e., a vector of random variables associated with message taking values in VL. • Each of the K foiling messages generated by the information foiling scheme for m can be thought of as a random variable vector taking values in VL. • We discussed three scenarios where different fake message generation schemes are designed with the performance requirements defined on the 3 metrics. • The scenarios are differentiated based on the foiler/attacker’s knowledge on the pmf for Vm:
Evaluation - methodology • Pub/sub service: stock quoting • Stock price volatility is a random walk with variance a normal distribution [Black-Scholes model] • Fake message generation: • Sit = St + ni , where Sit is the i-th fake message for the authentic stock price information St, and ni is white Gaussian noise. • Attacker’s strategy: • Uniform Sampling: The attacker picks each of the K+1 messages as the correct message with the same probability. • Extended Kalman Filter : Use an extended Kalman filter to generate estimates , and then picks the observed message j which is • data trace: • finance.yahoo.com
Evaluation results - 1 • The curve labeled “Sig. Events” shows the probability of correct guess by the attacker when the stock price changes by a large amount.
Evaluation results - 2 • A value of “Factor-10” means the variance of the noise was 10 times the variance of stock price. • higher variance for the added noise achieves a higher truth deviation.
Conclusion and Future Work • We propose a security mechanism called “information foiling” to address new confidentiality problems arising in pub/sub overlay services. • Information foiling extends Rivest’s ”Chaffing and Winnowing” idea. • Our scheme is complementary to the traditional cryptography-based security schemes and offers probabilistic guarantees on information confidentiality. • Many interesting open problems for future work. • The need for a stronger guiding theory to better understand • An analytic study on the fundamental trade-off between the fake message number, indistinguishability, and truth deviation is important. • Investigating the interaction between a foiler and an attacker in game theory. • The designs of optimal FMG schemes for other interesting and important application scenarios are needed.
Thank you! Questions?