SenShare: Transforming Sensor Networks into Multi-application Sensing Infrastructures

Ilias Leontiadis, Christos Efstratiou, Cecilia Mascolo, Jon CrowcroftEWSN 2012 · DOI

TL;DR

This paper introduces SenShare, a platform that transforms single-purpose sensor networks into shared infrastructures capable of supporting multiple co-existing applications. By decoupling hardware from applications and creating virtual overlay networks, SenShare enables efficient resource utilization and the deployment of diverse sensing applications without disrupting existing ones.

What Problem Does This Paper Address?

Traditional sensor networks are typically built for a single application and owned by a single entity. This approach becomes inefficient and costly as the demand for sensing data grows and more applications emerge. Deploying new applications often requires replicating infrastructure or complexly fusing multiple applications into a single binary, which is challenging when different authorities own the infrastructure and applications. The paper addresses the need for a flexible, shared sensing infrastructure that can be accessed by multiple applications and developers, particularly in environments like smart buildings where diverse sensing modalities are already present but underutilized.

Methods

SenShare tackles the challenge of shared sensing through several key innovations:

  • In-Node Hardware Abstraction Layer (HAL): Each sensor node has a HAL that acts as a unified interface for multiple applications to access hardware resources. This breaks the tight coupling between applications and specific hardware, allowing applications to be developed independently of the underlying physical devices.
  • Runtime Environment: A "Node Runtime" component resides on each sensor node, managing access to hardware and providing an abstraction layer. This runtime operates as a user-space process, interacting with dynamically loaded applications.
  • Virtual Overlay Networks: SenShare creates virtual overlay networks for each application on top of the physical infrastructure. This allows applications to perceive the network as their own dedicated sensor network, isolated from others.
  • Traffic Isolation: Application packets are augmented with an "application routing header" containing the application ID, sequence number, origin, and destination. This header enables the runtime to reconstruct original addresses and manage traffic isolation.
  • Overlay Formation Protocol: To connect potentially isolated clusters of application instances, SenShare uses a lightweight protocol that leverages the underlying infrastructure's routing tree (CTP). It identifies "cluster edges" and establishes virtual links to connect these clusters, ensuring all applications can reach the network sink.
  • Selective Control & Deployment: The platform provides tools for infrastructure owners to manage applications, including starting/stopping them, and enabling/disabling device access for specific applications. Deployment uses a modified version of the Deluge protocol.

Datasets

Not discussed in the paper. The evaluation relies on a testbed deployment and simulated scenarios.

Key Insights and Ideas

  • Decoupling is Key: The core idea is to separate the sensing infrastructure from the applications that use it, enabling a more flexible and scalable model.
  • Virtualization for Sharing: Creating virtual sensor networks on top of a physical infrastructure is a viable approach to enable multi-tenancy in sensor networks.
  • Overhead vs. Benefit: While introducing abstraction layers and runtimes incurs some performance overhead, this is a necessary trade-off for the significant benefits of sharing and multi-application support. The paper shows that optimizations can mitigate this overhead.
  • Leveraging Existing Infrastructure: SenShare cleverly reuses existing routing protocols (like CTP) for infrastructure management and overlay formation, minimizing the need for entirely new protocols.
  • Smart Buildings as a Prime Use Case: The paper highlights smart buildings as an ideal environment for shared sensing due to the prevalence of similar sensing modalities and the potential for diverse applications.

Key Results

  • Runtime Performance: The virtualized applications experienced a maximum sampling rate reduction of up to 28% compared to native applications. However, with multiple applications running concurrently, optimization mechanisms reduced this impact, demonstrating improved overall system performance through in-node coordination. CPU and memory usage showed a linear increase with the number of applications, which is considered a manageable price for sharing.
  • Network Overlay Performance: The overlay management traffic was found to be negligible (less than 0.1% of total traffic), especially after the initial exchange of distance vectors. This indicates a low overhead for maintaining overlay topologies.
  • Overlay Traffic Cost: The overhead for relaying packets between disjointed overlay clusters was higher for smaller deployments but dropped significantly as nodes formed larger clusters. The SenShare overlay mechanism showed significant gains over a simple collection mechanism by forwarding traffic only to the closest overlay cluster.
  • Overlay Topology: The overlay topology's weight varied, and while not always producing the absolute optimal topology (compared to a Minimum Spanning Tree), it performed better as the number of deployed nodes increased. The protocol offers a light-weight approach for establishing overlay networks.
  • Deployment: SenShare was successfully deployed in an office environment supporting environmental monitoring and office occupancy applications, demonstrating its real-world applicability.

Limitations and Future Work

  • Security: The authors acknowledge that significant and potentially novel security measures are needed to protect a shared network supporting applications from different users. Future work will investigate new classes of threats and expand SenShare with additional security mechanisms.
  • Hardware Abstraction: While SenShare offers advanced features, it relies on existing virtualization frameworks and aims to be complementary. The paper notes that for smaller nodes, multiprocessing real-time kernels could offer more task-running capabilities.
  • SQL-like APIs: The paper mentions that while SQL-like APIs can be sufficient for some applications, SenShare aims to support applications requiring specific sensing functionality and custom code on sensor nodes.
  • Optimal Topology: The overlay protocol does not always guarantee the most optimal topology, but it offers a good trade-off between discovery/maintenance cost and performance. Future work could explore more sophisticated topology discovery mechanisms.

Summary updated

BibTeX
@inproceedings{leontiadis2012senshare,
  author = {Ilias Leontiadis and Christos Efstratiou and Cecilia Mascolo and Jon Crowcroft},
  title = {{SenShare: Transforming Sensor Networks into Multi-application Sensing Infrastructures}},
  booktitle = {EWSN},
  year = {2012},
  doi = {10.1007/978-3-642-28169-3\_5},
}
Figure