Wireless Sensor Networks (WSNs) are networks consisting of small, battery-powered computers with short-range radio communication and sensing capabilities. These computers (referred to as nodes) are used to sense one or more variables using one or more sensors and report these readings to a base-station via a multi-hop communication. Often, these WSNs are deployed to detect a phenomenon. Detection of this phenomenon usually depends on readings from several sensors in different locations. Therefore, sensor readings are periodically collected at the base-station which processes these data or forwards them to a cloud. This base-station also represents a gateway for users to access and communicate with the WSN. It allows a user to submit a query, whose execution retrieves data from relevant sensor nodes and the result of the computation over these data is detection of a phenomenon. In a typical node, radio is responsible for far more energy consumption when compared to the CPU or most of the sensors. Therefore, it has always been researchers’ intention to lower the network communication to the lowest possible level. Because nodes closer to the base-station transfer more data, their batteries are depleted faster which may lead to part of the network being unreachable. Additionally, because a user accesses the WSN via a base-station, it represents a single point of failure. One of the solutions to overcome this problem is to allow a user to communicate and submit a query via any node in the network. However, building a fully decentralised and energy-efficient framework allowing any node to accept and execute a query submitted by a user brings several new challenges. First, a node needs to be able to communicate with any other node in the network, not only the base-station, without relying on any central entity. Second, any node must be able to identify all the nodes which monitor the same phenomenon. And third, a node which processes the data must be chosen in such way, that the overall communication of the whole network is minimised.
In this thesis we present Dragon, a framework for in-network data stream processing. Dragon allows communication among any pair of nodes via optimal or near optimal routes. This is achieved without the need to first discover or establish a path between two communicating nodes. Dragon also allows any node to find a list of all other nodes fulfilling given static criteria. The search for these nodes requires communication with only close (possibly multi-hop) neighbourhood. Finding a list of nodes observing the same phenomenon and requesting data directly from these nodes allows any node in the network to accept and execute a snapshot iii(one-time) query with a very low network overhead and in a timely manor. Finally, Dragon introduces a distributed algorithm for discovery of a processing node for continuous queries in WSNs. The algorithm follows the cost gradient to the node with the lowest communication cost, hence decreasing the overall network traffic and communication delay.