Data collection is one of the major traffic pattern in wireless sensor networks, which requires regular source nodes to send data packets to a common sink node with limited end-to-end delay. However, the sleep latency brought by duty cycling mode results in significant rise on the delivery latency. In order to reduce unnecessary forwarding interruption, the state-of-the-art has proposed pipeline scheduling technique by allocating sequential wakeup time slots along the forwarding path. We experimentally show that previously proposed pipeline is fragile and ineffective in reality when wireless communication links are unreliable. To overcome such challenges and improve the performance on the delivery latency, we propose Robust Multi-pipeline Scheduling (RMS) algorithm to coordinate multiple parallel pipelines and switch the packet timely among different pipelines if failure happens in former attempts of transmissions. RMS combines the pipeline features with the advantages brought by multi-parents forwarding. Large-scale simulations and testbed implementations verify that the end-to-end delivery latency can be reduced by 40% through exploiting multi-pipeline scheduled forwarding path with tolerable energy overhead.

Working in the duty cycling mode enables sensor nodes to utilize limited energy efficiently instead of unnecessary idle listening. In such networks, awareness of neighboring nodes' working schedules is essential, especially when each node sets up the schedule independently. Most traditional research assumes that a node can always share its working schedule with its neighbors once it joins the network. However, dynamic energy supply and varied requirements of system performance make adjusting duty cycles necessary. Consequently, sensor nodes have to regularly change their schedules and advertise the new schedule to their one-hop neighbors. In this work, instead of changing nodes' working schedule directly, we introduce SSD, a staged and smooth design for safeguarding schedule updates. Our design is aimed at bounding average packet loss rate with a minimum energy cost. We evaluate our design with large-scale simulation and show that under the comparable packet loss rate, our design saves up to 35% energy compared with the minimal packet loss solution.