Distributed Joins in Sensor Networks
Svilen Mihaylov, Zack Ives, Sudipto Guha
University of Pennsylvania

As sensor networks become increasingly sophisticated, providing
a variety of multimodal inputs, a key operation will be
to integrate heterogeneous data across multiple devices and
device types, e.g., to combine RFID and video frame data
in order to record what is being monitored, or to match up
patients and caregivers. The fundamental primitive in such
efforts will be an in-network join of samples from different
collections of devices, based not only on physical properties
(e.g., proximity) but also logical ones (e.g., identifiers,
features).
In this paper, we study the problem of performing joins
over a sensor network, when the network is represented as a
set of tables that change over time. We develop techniques
for establishing communication to support in-network join;
placement of the join computation within the network; and
cost modeling and optimization in this context. Compared
to the naive approach of joining values at the root, our algorithms
provide significantly better power distribution, fault
tolerance, and performance. Our work represents a first step
towards integrating data from heterogeneous sensor types.