Approximate Query Processing in Sensor Networks

Abstract

Many emerging applications are collecting massive volumes of sensor data from networks of distributed devices, such as sensor networks and cyber-physical systems. These environments are commonly characterized by the intrinsic volatility and uncertainty of the sensor data, and the strict communication (energy) constraints of the distributed devices. Approximate query processing is an important methodology that exploits the tolerance of many applications to inaccuracies in reported data in order to reserve communication overhead. The research challenge is how to ensure communication efficiency without sacrificing result usefulness.

Many prior work depends on users to impose preferences or constraints on approximate query processing, such as result inaccuracies, candidate set size, and response time. We argue that the pre-determined user preferences may turn out to be inappropriate and become a substantial source of i) jeopardized query results, ii) prohibitive response time, and iii) overwhelming communication overhead.

Our work ‘Probing Queries in Wireless Sensor Networks’ (ICDCS 2008) studies a scenario where empty sets may be returned as accurate query results, yet users may benefit from approximate answer sets not exactly conforming the specified query predicates. The approximate answer sets can be used not only to answer the query approximately but also to guide users to modify their queries for further probing the monitored objects. The distance between sensing data and a query and the dominating relationship between sensing data are first defined. Then, three algorithms for processing probing queries are proposed, which compute the best approximate answer sets that consist of the sensing data with the smallest distance from given queries. All the algorithms utilize the dominating relationship to reduce the amount of data transmitted in sensor networks by filtering out the unnecessary data. Experimental results on real and synthetic data sets show that the proposed algorithms have high performance and energy efficiency.

Our work ‘Enabling ε-Approximate Querying in Sensor Networks’ (VLDB 2009) studies the scenario where, due to the dynamic nature of sensor data, users are unable to determine in advance what error bounds can lead to affordable cost in approximate query processing. We propose a novel ε-approximate querying (EAQ) scheme to resolve the problem. EAQ is a uniform data access scheme underlying various queries in sensor networks. The core idea of EAQ is to introduce run-time iteration and refinement mechanisms to enable efficient, ε-approximate query processing in sensor networks. Thus it grants more flexibility to in-network query processing and minimizes energy consumption through communicating data up to a just-sufficient level. To ensure bounded overall cost for the iteration and refinement procedures of the EAQ scheme, we develop a novel data shuffling algorithm. The algorithm converts sensed datasets into special representations called MVA. From prefixes of MVA, we can recover approximate versions of the entire dataset, where all individual data items have guaranteed error bounds. The EAQ scheme supports efficient and flexible processing of various queries including spatial window query, value range query, and queries with QoS constraints. The effectiveness and efficiency of the EAQ scheme are evaluated in a real sensor network testbed.

Even in case the users know exactly what result inaccuracies they can tolerate, many prior query processing techniques still cannot meet arbitrary precision requirements given by users. Most notably, many aggregational query processing methods can only support fixed error bounds.

In ‘Sampling based (ε, δ)-Approximate Aggregation Algorithm in Sensor Networks’ (ICDCS 2009), we propose a uniform sampling based aggregation algorithm. We prove that for any ε(ε ≤ 0) and δ(0 ≤ δ ≤ 1), this algorithm returns the approximate aggregation result satisfying that the probability of the relative error of the results being larger than ε is less than δ. However, this algorithm is only suitable for static network. Considering the dynamic property of sensor networks, we further proposed a Bernoulli sampling based algorithm in ‘Bernoulli Sampling based (ε, δ)-Approximate Aggregation in Large-Scale Sensor Networks’ (INFOCOM 2010). We prove that this algorithm also can meet the requirement of any precision, and is suitable for both static and dynamic networks. Besides, two sample data adaptive algorithms are also provided. One is to adapt the sample with the varying of precision requirement. The other is to adapt the sample with the varying of the sensed data in networks. The theoretical analysis and experiments show that all proposed algorithms have high performance in terms of accuracy and energy cost.