Simulatable Binding: Beyond Simulatable Auditing

Abstract

A fundamental problem in online query auditing is that an outside attacker may compromise database privacy by exploiting the sequence of query responses and the information flow from the database state to the auditing decision. Kenthapadi et al. [14] proposed the simulatable auditing model to solve this problem in a way that completely blocks the aforementioned information flow. However, the security does not come for free. The simulatable auditing model actually suffers from unnecessary data utility loss.

We assert that in order to guarantee database privacy, blocking the information flow from the true database state to the auditing decision is sufficient but far from necessary. To limit the loss in data utility, we suggest an alternative approach that controls, instead of blocks, such information flow. To this end, we introduce a new model, called simulatable binding, in which the information flow from the true database state to the auditing decision is provably controlled by a selected safe binding. We prove that the proposed simulatable binding model provides a sufficient and necessary condition to guarantee database privacy, and therefore, algorithms based on our model will provide better data utility than algorithms based on the simulatable auditing model. To demonstrate the strength and practicality of our model, we provide two efficient algorithms for the max query and sum query auditing, respectively. For the ease of comparison, each algorithm is built by applying our simulatable binding model, and is compared to an algorithm applying the simulatable auditing model. Clear improvements are shown through experiments.