Large-scale Network Discovery: Design Tradeoffs in Wireless Sensor Systems

We are investigating design tradeoffs relating to the structure of the network formed in large-scale selforganizing wireless sensor systems. In such systems, most traffic is in the form of many-to-one data flows, between all the networked sensors and a central monitoring node. We refer to the process of establishing a multi-hop routing path from every sensor node in the system to the monitoring node as Network Discovery. Flooding, which is conventionally used for disseminating information, can also be used for network discovery. The central monitoring node initiates the flooding by forwarding a message to all its neighbors, each successive node forwarding it once. As the flooding proceeds, nodes keep track of who they received messages from. If a given node receives multiple copies of the flooding message from different neighboring senders, it may use a parent-selection mechanism to choose one of these as its parent. The nodes then send information to their parent, along along the reverse direction of the original flooding message, when they need to forward any data back to the monitoring node. The network topology that is thus discovered is a spanning tree. It is of interest to know whether the discovered routing tree is bushy, with a few large clusters, or sparse, with many smaller clusters. Three factors affecting the structure of the discovered routing tree are the flooding mechanism, the transmission power, and the parent selection algorithm. The bushiness of the discovered tree structure will have direct implications in terms of application data aggregation, energy utilization, system throughput and robustness. We would like to address a number of related questions in our investigations. What are the tradeoffs among these implications? Increasing the radio transmission power decreases the degree of spatial reuse in the network, but results in a bushier tree structure where greater application data aggregation can occur. How does this change the overall system throughput? Bushy trees also result in highly non-uniform energy utilization as opposed to sparse trees; but do they utilize less energy overall? How robust are bushy trees to different kinds of node failures? Finally, how can the use of different parent selection mechanisms help in controlling the bushiness of the discovered tree? Our initial results in this study are obtained from a combination of theoretical models and empirical data from prototype wireless sensor networks consisting of over 150 nodes. These results suggest that when the transmission radius of nodes is large, and basic flooding with a simple parent-selection mechanism is used, the topology of the discovered routing tree is in fact bushy and follows a power law distribution. Most nodes are leaf nodes, and a small-yet-significant number of the nodes are cluster-heads with a large number of children. This happens because , in a given area, the first node which wins the medium-access contention and broadcasts the flood message first becomes the parent of all new receivers. This “winner gets all” scenario is