Duty-Cycle Distribution in Low-Power Collection

We characterize the interplay between collection and lowpower operation by studying the joint impact of the offered load and the Low Power Listening sleep time setting on the duty-cycle distribution in a large-scale sensor network testbed. Collection services must be able to work on top of a low-power MAC. The Collection Tree Protocol (CTP) [1] has been shown to perform very well on top of Box-MAC [2], the mainstream TinyOS MAC layer that provides Low Power Listening (LPL) [3] support: CTP achieves duty-cycles of 3% with an InterPacket Interval (IPI) of 5 minutes and an LPL sleep interval of 1s. Another tree-based collection protocol for low-end sensor networks, Arbutus, has been recently shown to compare favorably with CTP in a very extensive analysis [4] that is, however, limited to an always-on CSMA MAC layer. Our ongoing work has a twofold goal: (1) to validate Arbutus in the presence of MAC-layer dutycycling; (2) to explore the impact of the IPI and the sleep time settings on the duty cycle statistics. The motivation for (1) is clear: without duty-cycling, the radio is a toy. The motivation for (2) lies in the lack of a systematic study about the effect of different sleep time settings as a function of the offered load. It is well-accepted that nodes should sleep, but how often should they wake up given the extent of their projected load? This work presents our progress toward these two goals. Sleep Time [ms] IPI [min] Duty Cyle [%] (μ± σ) Average Delivery Rate 250 1 10.31± 3.39 0.97 1000 1 11.63± 9.04 0.93 250 5 4.81± 2.59 0.98 1000 5 3.16± 1.04 0.98 Table 1. Overall results: duty-cycle and delivery rate at four different operating points. We adapted the TinyOS implementation of Arbutus used in [4] for LPL support, and we tested it on top of BoX-MAC on the Motelab testbed [5]. We fixed the topology by focusing on a specific sink assignment (node 103) that the results in [4] showed to be particularly favorable. Table 1 shows the results with about 100 Motelab nodes, averaged over five 30min runs for four different IPI-Sleep operating points. The results at IPI=5min compare favorably with the 0 5 10 15 20 25 0 10 20 30 40 50 60 70 80 90 100