Developing Pervasive Agent-Based Applications: A Comparison of Two Coordination Approaches

Inmaculada Ayala, Mercedes Amor, Lidia Fuentes, Marco Mamei, Franco ZambonelliAose 2012 · DOI

TL;DR

This paper compares two distinct agent-based coordination approaches – FIPA-compliant messaging (MalacaTiny-Sol) and tuple spaces (SAPERE) – for developing pervasive applications. By modeling scenarios in an Intelligent Museum, the authors provide a detailed evaluation of their strengths and weaknesses across various software engineering and pervasive computing metrics, offering insights into choosing the right paradigm for specific application needs.

What Problem Does This Paper Address?

Pervasive computing aims to create environments that are sensitive, adaptive, and responsive to human needs, often leveraging interconnected devices. Agent technology is well-suited for this due to agents' autonomy, reactivity, and social capabilities. However, there's no dominant approach for coordinating these agents in pervasive systems. This paper addresses this by comparing two major coordination paradigms: FIPA interaction protocols (direct communication) and tuple spaces (indirect communication), to understand their trade-offs in building such complex systems.

Methods

The authors propose a comparative evaluation of two agent systems: MalacaTiny-Sol, representing FIPA-compliant coordination, and SAPERE, representing tuple-space-based coordination. They model several scenarios within an Intelligent Museum (IM) context, including information provision and emergency evacuation. The evaluation uses a combination of software engineering metrics (Separation of Concerns, Coupling and Cohesion) and pervasive computing concerns (Adaptability, Robustness, Scalability, Privacy).

Datasets

Not discussed in the paper. The evaluation uses simulated scenarios within an Intelligent Museum context.

Key Insights and Ideas

  • Coordination is Key: The choice between direct (FIPA) and indirect (tuple spaces) coordination significantly impacts agent design, system complexity, and overall performance in pervasive environments.
  • Separation of Concerns (SoC): SAPERE excels in architectural separation at the infrastructure level (CDAC, CDAI), while MalacaTiny-Sol is better at separating concerns within the agent itself (CDAO) through its aspect-oriented approach.
  • Adaptability vs. Infrastructure: MalacaTiny-Sol offers more flexibility in adapting agent architecture and behaviors, while SAPERE's distributed tuple space infrastructure provides inherent adaptability to physical space changes.
  • Privacy Trade-offs: MalacaTiny-Sol generally offers better privacy by keeping computation local and minimizing data sharing, whereas SAPERE's distributed nature can lead to more information being accessible.
  • Hybrid Potential: The paper suggests that combining the strengths of both approaches (e.g., deploying MalacaTiny agents on SAPERE's robust infrastructure) could yield even more powerful pervasive systems.

Key Results

  • SoC: SAPERE scored better for Concern Diffusion over Architectural Components (CDAC) and Interfaces (CDAI), indicating better infrastructure-level separation. MalacaTiny-Sol scored better for Concern Diffusion over Operations (CDAO), reflecting its fine-grained aspect-based separation within agents.
  • Coupling and Cohesion: MalacaTiny-Sol showed better cohesion (higher LCC) and higher fan-out, suggesting better internal modularity and reusability of agent components. SAPERE had better fan-in, indicating fewer dependencies.
  • Adaptability: MalacaTiny-Sol requires more effort to adapt to changes in physical space but offers more agent-level behavioral adaptation. SAPERE's infrastructure handles physical space changes more gracefully, with less agent modification needed.
  • Robustness: MalacaTiny-Sol's distributed functionality between agents and the AP provides better application-level robustness. SAPERE's distributed nodes offer a more robust infrastructure than Sol's centralized AP.
  • Scalability: SAPERE agents are simpler and require fewer elements, but MalacaTiny-Sol offers better component reuse and a more uniform extension mechanism for agent capabilities.
  • Privacy: MalacaTiny-Sol generally provides better privacy as users share less information and computation is more localized.

Limitations and Future Work

  • Limitations: The evaluation is based on specific case studies within an Intelligent Museum. The robustness of Sol's infrastructure is noted as a potential weakness, though it acknowledges that other FIPA approaches might be more robust. The paper also points out that SAPERE's distributed nature can make it harder to control the number of people directed to different exits in emergency scenarios.
  • Future Work: The authors explicitly state their plan to study the combination of these approaches, suggesting that integrating MalacaTiny's aspect-oriented agent design with SAPERE's robust, distributed tuple space infrastructure could be a promising direction.

Summary updated

BibTeX
@inproceedings{ayala2012developing,
  author = {Inmaculada Ayala and Mercedes Amor and Lidia Fuentes and Marco Mamei and Franco Zambonelli},
  title = {{Developing Pervasive Agent-Based Applications: A Comparison of Two Coordination Approaches}},
  booktitle = {Aose},
  year = {2012},
  doi = {10.1007/978-3-642-39866-7\_5},
}
Figure