Secure and highly-available aggregation queries in large-scale sensor networks via set sampling

Wireless sensor networks are often queried for aggregates such as predicate count, sum, and average. In untrusted environments, sensors may potentially be compromised. Existing approaches for securely answering aggregation queries in untrusted sensor networks can detect whether the aggregation result is corrupted by an attacker. However, the attacker (controlling the compromised sensors) can keep corrupting the result, rendering the system unavailable. This paper aims to enable aggregation queries to tolerate instead of just detecting the adversary. To this end, we propose a novel tree sampling algorithm that directly uses sampling to answer aggregation queries. It leverages a novel set sampling technique to overcome a key and well-known obstacle in sampling - traditional sampling technique is only effective when the predicate count or sum is large. Set sampling can efficiently sample a set of sensors together, and determine whether any sensor in the set satisfies the predicate (but not how many). With set sampling as a building block, tree sampling can provably generate a correct answer despite adversarial interference, while without the drawbacks of traditional sampling techniques.