Poster Abstract: If You Have Time, Save Energy with Pull

David Hasenfratz, Andreas Meier, Matthias Woehrle, Marco Zimmerling, Lothar Thiele

Abstract

In wireless sensor networks (WSNs), the traditional "push" approach—where nodes send data immediately—wastes significant energy maintaining active routing paths. This paper presents a "pull" approach that buffers data in local flash memory and transmits it in infrequent bursts. Experimental results from a 35-node testbed demonstrate up to a 50% reduction in power consumption when applications can tolerate higher latency.

Introduction

Most wireless sensor networks are battery-powered and intended for long-term deployment, making energy efficiency the primary design constraint. Traditionally, these networks use a "push" model: as soon as a node senses data, it tries to route it to a central sink. To make this work, nodes must stay relatively active to maintain up-to-date routing trees, even when they aren't sending much data.

For many applications, such as environmental monitoring, immediate data delivery is not strictly necessary. This research explores trading latency for battery life by transitioning from a push-based system to a pull-based one. The core insight is that latency is a resource that can be traded for energy; if data is not needed immediately, the network does not need to remain "warm" and active.

Methodology

The researchers transformed the standard Collection Tree Protocol (CTP) into a pull-based system by introducing two distinct phases:

  1. Sleep Phase: Nodes store sensed data in local flash memory. They do not communicate, allowing the radio to stay in a low-power state with a long MAC wake-up interval.
  2. Data Collection Phase: The sink initiates this phase by broadcasting a beacon. Nodes then initialize routing paths, switch to a shorter MAC wake-up interval to increase bandwidth, and "dump" their backlogged flash data to the sink.

To ensure reliability, the system implements ack-based deletion and back-pressure flow control. A key finding is that the energy required to write to and read from local flash memory is negligible compared to the energy consumed by the radio, particularly during idle listening.

Power consumption breakdown by protocol phase, highlighting the dominance of idle listening and the low cost of flash operations.

Key Results

The pull approach was evaluated on a physical 35-node testbed of Tmote Sky devices. The results indicate a clear "break-even" point: the pull approach becomes more energy-efficient than the push approach once the pull interval exceeds 5 minutes.

Comparison of power consumption between Push and Pull protocols. The Pull approach shows significant energy savings as the pull interval increases.

With a pull interval of 60 minutes, the average power consumption dropped by 50%, and the maximum power consumption (at the bottleneck node) dropped by 40% compared to the push approach. Nodes consumed 90% less power during the sleep phase than during the data collection phase, primarily due to reduced idle listening. Despite the bursty nature of the pull approach, the reliability modifications allowed for 100% data yield in experiments.

Discussion and Limitations

The primary limitation of this approach is its unsuitability for time-critical applications like fire detection or industrial alarms. Additionally, the maximum pull interval is bounded by the size of the node's flash memory; exceeding this limit could lead to data loss or require emergency uploads. Future work could involve performing data compression or aggregation on the stored "chunks" before transmission to further optimize energy usage.

Summary updated

Abstract

We analyze push and pull for data collection in wireless sensor networks. Most applications to date use the traditional push approach, where nodes transmit sensed data immedi- ately to the sink. Using a pull approach, nodes store the data in their local flash memory, and only engage in commu- nication during dedicated collection phases. We show how one can transform an existing push-based collection protocol into a pull-based one, and compare the power consumption of both approaches on a 35-node testbed. Our results show that substantial energy gains are possible with pull, provided that the application can tolerate a long latency.
BibTeX
@inproceedings{hasenfratz2010poster,
  author = {David Hasenfratz and Andreas Meier and Matthias Woehrle and Marco Zimmerling and Lothar Thiele},
  title = {{Poster Abstract: If You Have Time, Save Energy with Pull}},
  booktitle = {SenSys},
  year = {2010},
  url = {https://openreview.net/forum?id=CNKoGon3UME},
}
Figure