Self-adaptive software needs quantitative verication at runtime

Radu Călinescu, Carlo Ghezzi, Marta Kwiatkowska, Raffaela MirandolaHAL 2012

TL;DR

This paper argues that to achieve dependable self-adaptive software, formal verification techniques like quantitative verification must be integrated into the runtime adaptation process. They propose a framework where quantitative verification is used to monitor, analyze, plan, and execute adaptations, ensuring that evolving software continues to meet its requirements.

What Problem Does This Paper Address?

Modern software systems are increasingly expected to be both adaptive (able to change and evolve in response to their environment) and dependable (reliable and safe, especially in critical applications). Traditionally, these two concerns have been addressed by separate research communities: autonomic computing for adaptation and formal methods for dependability. This paper tackles the challenge of bridging this gap. The core problem is how to ensure that as software adapts itself at runtime, it doesn't compromise its dependability. This is particularly crucial in environments with high uncertainty and variability, where software must continuously adjust to maintain its objectives.

Methods

The paper proposes integrating quantitative verification techniques into the MAPE (Monitor-Analyze-Plan-Execute) loop of autonomic computing.

  • Quantitative Verification: This is a mathematically-based technique used to establish the correctness, performance, and reliability of systems with stochastic behavior. It involves creating a formal model of the software system and its requirements, then analyzing the model's compliance.
  • Runtime Integration: The authors suggest using quantitative verification at various stages of the adaptation process:
  • Monitoring: Augmenting the system with a component that continuously updates a quantitative model of the system based on observed behavior. This can involve using techniques like Bayesian learning for failure rates or Kalman filters for performance parameters.
  • Analysis: Automatically invoking a quantitative verification tool to detect or predict requirement violations by checking the updated model against formally specified requirements.
  • Planning: Using quantitative verification to suggest adaptive maintenance steps that ensure the system continues to satisfy its requirements after changes. This can involve selecting optimal configurations or services.
  • Formalism: The paper uses a framework where requirements (R) are satisfied by the specification (S) in the context of domain assumptions (D), expressed as $S, D \models R$. Quantitative verification is used to check this relationship at runtime.

Datasets

Not discussed in the paper. The paper focuses on a conceptual framework and uses a simplified medical assistance system as an illustrative example.

Key Insights and Ideas

  • Bridging the Divide: The most significant insight is the necessity of merging formal verification with self-adaptation. Simply having adaptive systems isn't enough; they must be dependably adaptive.
  • Runtime Formal Verification is Key: Traditional offline verification is insufficient for systems that need to adapt continuously. Formal verification must become an active, runtime component.
  • MAPE Loop Enhancement: Quantitative verification can provide the rigorous foundation needed for each stage of the MAPE loop, moving beyond ad-hoc adaptation to a more systematic and provably correct approach.
  • Predictive Power: Quantitative verification can not only detect current violations but also predict future ones, allowing for proactive adaptation.
  • Formal Guarantees for Adaptation: The ultimate goal is to provide irrefutable proofs that the reconfigured software will continue to meet its requirements, even as it evolves.

Key Results

The paper doesn't present empirical results from a large-scale deployment but rather outlines a framework and demonstrates its potential through a simplified medical assistance system example. The key "results" are conceptual:

  • A methodology for integrating quantitative verification into the runtime adaptation of software.
  • Demonstration of how this integration can support monitoring, analysis, and planning stages of adaptation.
  • Illustrative examples showing how quantitative verification can identify requirement violations and guide the selection of alternative system configurations (e.g., choosing different alarm services).
  • The paper highlights that this approach can augment software systems with self-adaptation capabilities in predictable ways.

Limitations and Future Work

  • Scalability: While the paper mentions research into faster runtime verification algorithms for larger systems, the overhead of quantitative verification at runtime can still be a concern for very large or highly dynamic systems.
  • Domain Dependence: The effectiveness of adaptation strategies is acknowledged to be domain and application-dependent, suggesting that a catalog of strategies might be needed.
  • Model Evolution: The paper touches upon the need for models to be "kept alive" at runtime and the challenges of deriving implementations from models dynamically.
  • New Execution Platforms: The rise of cloud computing introduces new challenges, as the infrastructure itself can change. Future work needs to consider "holistic" solutions that adapt both the application and the underlying platform.
  • Discovery and Model Learning: A significant area for future research is how components can learn about each other's capabilities in dynamic, service-rich environments, especially with black-box visibility.
  • Effective Integration: Developing a repertoire of techniques for timely and effective reactions to detected violations, and exploring model-driven approaches for self-adaptation, are crucial future directions.

Summary updated

Abstract

International audience
BibTeX
@inproceedings{calinescu2012self,
  author = {Radu Călinescu and Carlo Ghezzi and Marta Kwiatkowska and Raffaela Mirandola},
  title = {{Self-adaptive software needs quantitative verication at runtime}},
  booktitle = {HAL},
  year = {2012},
}
Figure