Developing and deploying sensor network applications with AnduIN

Daniel Klan, Katja Hose, Kai-Uwe SattlerDMSN 2009 · DOI

TL;DR

This paper introduces AnduIN, a system that bridges the gap between centralized data stream management (DSMS) and in-network query processing (INQP) for wireless sensor networks (WSNs). AnduIN offers an optimizer to intelligently distribute query execution between the resource-constrained sensor nodes and a powerful central engine, aiming to minimize energy consumption and extend network lifetime.

What Problem Does This Paper Address?

Wireless sensor networks (WSNs) are increasingly used for applications like environmental monitoring and traffic analysis. However, these sensors are battery-powered, making energy efficiency paramount. Sending all raw sensor data to a central processing unit is energy-prohibitive. While in-network query processing (INQP) can reduce communication overhead by processing data closer to the source, it often lacks the power for complex data analysis. Existing systems either don't support sophisticated data mining tasks or lack a declarative approach for query formulation and deployment. This paper addresses the challenge of developing and deploying WSN applications that can handle complex data analysis while optimizing for energy efficiency, by proposing a unified system that combines DSMS and INQP capabilities.

Methods

AnduIN employs a multi-component architecture:

  • DSMS (AnduIN Core): A query processor that handles queries formulated in CQL (Continuous Query Language), supporting standard operators like filters, joins, and aggregations, as well as "synopsis operators" for complex data mining tasks (e.g., burst detection).
  • In-Network Query Processors (INQPs): A library of C functions running on sensor nodes (specifically on Contiki OS) that implement basic query operators.
  • Web-based Frontend: A user-friendly graphical interface (inspired by Yahoo! Pipes) that allows users to visually design queries using a box-and-arrow paradigm. This graphical representation is then translated into CQL queries.
  • Optimizer: A crucial component that decomposes query plans, deciding which parts of a query should be executed in-network and which on the central DSMS. This optimizer considers complex data analysis tasks and aims to minimize overall network power consumption.

Key novelties include:

  • Declarative Query Specification: A flexible approach to define how sensor data should be processed, separating data modeling from data manipulation.
  • Query Decomposition and Deployment: Automatic translation of high-level queries into executable plans for both the DSMS and the sensor nodes, including over-the-air programming for sensor nodes.
  • Cost-Based Optimization: A physical optimizer that enumerates possible query execution plans and selects the one that minimizes estimated network power consumption, considering factors like message costs, wake-up times, and operator selectivity.

Datasets

Not discussed in the paper. The paper focuses on a general framework and uses illustrative examples rather than specific datasets for evaluation.

Key Insights and Ideas

  • In-network processing isn't always the most power-efficient solution: The paper demonstrates through its cost model and case study that a hybrid approach, where some complex operations are still offloaded to the central DSMS, can be more energy-efficient depending on data characteristics and operator selectivity.
  • Intelligent query decomposition is key: The optimizer's ability to strategically split queries between the sensor network and the central DSMS is critical for balancing computational complexity with energy constraints.
  • Usability matters for WSN application development: The graphical query formulation frontend significantly lowers the barrier to entry for users, making complex WSN application development more accessible.
  • Synopsis operators enable complex analysis on resource-constrained devices: The introduction of synopsis operators allows for sophisticated data mining tasks to be integrated into the query processing pipeline, even on sensor nodes.

Key Results

The paper presents a conceptual framework and an architecture for AnduIN. While quantitative results are not the primary focus, the paper:

  • Illustrates the query decomposition process with an example for anomaly region detection.
  • Provides a cost model to estimate power consumption for different query execution plans.
  • Shows through a case study and accompanying figures (Figure 6) how power consumption varies with operator selectivity, highlighting scenarios where in-network processing is advantageous and others where it is not.
  • Demonstrates the practical implementation of the system with a Web-based GUI for query formulation.

Limitations and Future Work

  • Current cost model limitations: The paper acknowledges that the cost model relies on estimations, especially when statistics for data distribution are unavailable. Future work aims to improve these estimations.
  • Optimization criteria: Currently, the primary optimization goal is overall network power consumption. Future work plans to incorporate other criteria like data throughput and quality of service.
  • Dynamic data distribution: The paper notes that if data distribution changes, running query plans might need reorganization. This dynamic adaptation is a direction for future research.
  • Bulk registration: While planned, the XML-based sensor model language for bulk registration of sensors is not fully implemented in the presented prototype.

Summary updated

BibTeX
@inproceedings{klan2009developing,
  author = {Daniel Klan and Katja Hose and Kai-Uwe Sattler},
  title = {{Developing and deploying sensor network applications with AnduIN}},
  booktitle = {DMSN},
  year = {2009},
  doi = {10.1145/1594187.1594202},
}
Figure