Resilient Data Aggregation and Aggregator Node Election in Sensor Networks
Péter Schaffer · 2009
Wireless sensor networks are considered as a promising technology that has a wide range of applications including environmental monitoring for agricultural and ecological purposes, wild life monitoring, remote patient monitoring in electronic health care systems, building automation, and reconnaissance applications for military purposes, etc. Sensor networks typically consist of a large number of sensor nodes and a few base stations. The sensor nodes measure some physical phenomena (e.g., temperature, humidity, vibration) that are important in the given application, and report their sensor readings to the base stations (typically via wireless communication channels). As both the number of the sensors and the amount of the measurements that they perform can be large, in many applications, the base stations or some intermediate nodes aggregate the individual sensor readings into a compact report. Aggregation can be useful to keep the amount of information that need to be handled under control, and to improve the energy-efficiency of the network. A potential problem in this scenario is that sensor readings can be compromised before they reach the base station or the aggregator node. This can be achieved by an attacker for example by altering the environmental parameters around some sensors and thus corrupt their readings. This type of attack cannot be detected, nor prevented, by cryptographic means. In addition, this type of attack is relatively easy to carry out: Firstly, an attacker can easily approach a sensor node, as sensor networks are typically assumed to operate in an unattended manner. Secondly, corrupting the measurement of a nearby sensor does not require sophisticated mechanisms, but in most of the cases, everyday tools can be used effectively (e.g., a lighter, a pocket lamp, or a glass of water can be used to corrupt temperature, light, and humidity measurements, respectively). Unfortunately, many useful aggregation functions are sensitive to even a single compromised sensor reading. In my dissertation, I propose countermeasures against this type of attack – i.e., CORA and RANBAR – that are based on statistical hypothesis testing and sample filtering. The common property of the proposed solutions is that they perform an analyzing step before the aggregation, and with that, one is able to detect an attack or even filter out the compromised measurements. CORA is a twosample homogeneity test that is flexible enough to be applied without any special assumption on the distribution of the sensor readings or on the strategy of the attacker, while RANBAR is able to filter out a high percent of compromised measurement data by leaning on only one preassumption, namely that the sample is independent and identically distributed in the unattacked case. Besides protecting the aggregation function against input attacks, it is also highly important to efficiently assign the role of the aggregator among the nodes. As sensor nodes are usually resourceconstrained, and the aggregator nodes consume more energy than usual nodes (i.e., they spend