Scalability is an important criteria for sensor algorithm design. In this section, we evaluate scalability along two axes -- network size and the number of queries posed on a sensor network. Network size can vary depending on the application (e.g: the Extreme Scaling deployment [7] used 10,000 nodes, whereas the Great Duck Island deployment [14] used 100 nodes). The querying rate depends on the popularity of sensor data, for instance, during an event such as an earthquake, seismic sensors might be heavily queried while under normal circumstances, the query load can be expected to be light.
The testbed used in the scalability experiments comprises one Stargate proxy, twenty Telos mote sensor nodes, and an EmStar emulator that enables us to introduce additional virtual sensor nodes and perform larger scale experiments. Messages are exchanged between each sensor and the proxy through a multihop routing tree rooted at the proxy. Each sensor node is assumed to be operating at 1% duty-cycling. Since MAC layers that have been developed for the Telos mote do not currently support duty-cycling, we emulate a duty-cycling enabled MAC-layer. This emulator adds appropriate duty-cycling latency to each packet based on the microbenchmarks that we presented in Table 2.