Trustworthy Cloud Certification: A Model-Based Approach

Marco Anisetti, Claudio A. Ardagna, Ernesto Damiani, Nabil El IoiniSimpda (revised Selected Papers) 2014 · DOI

TL;DR

This paper proposes a novel model-based approach for trustworthy cloud certification that continuously verifies service behavior against a formal model at runtime. This addresses the challenge of maintaining certificate validity in dynamic cloud environments where services can change unexpectedly, ensuring ongoing trust between providers and customers.

What Problem Does This Paper Address?

Cloud computing offers immense flexibility and scalability, but its dynamic and opaque nature poses a significant challenge to establishing and maintaining trust. Customers need assurance that cloud services will behave as expected, especially concerning Quality of Service (QoS), while providers need to demonstrate the truthfulness of their claims. Traditional certification methods struggle in the cloud because services can change unpredictably after certification, potentially invalidating existing certificates. This paper tackles the problem of ensuring that cloud service certifications remain valid over time, even as the underlying services evolve.

Methods

The core of the proposed approach is model verification. The authors introduce a certification process involving a service provider, a certification authority, and an accredited lab. The accredited lab is responsible for:

  • Model Generation/Verification: Creating or verifying a formal model (Finite State Machine - FSM) of the cloud service (Target of Certification - ToC). This model can be either workflow-based (describing operation sequences) or implementation-based (including finer-grained details).
  • Execution Trace Collection: Gathering execution traces from the real ToC, either through monitoring actual customer interactions or by running synthetic test cases. These traces can represent sequences of operation calls (workflow level) or code instructions (implementation level).
  • Model Verification: Comparing the collected execution traces against the service model. This involves checking for inconsistencies such as:
  • Partial path discovery: A trace matches only a part of a model path.
  • New path discovery: A trace reveals a behavior not present in the model.
  • Broken existing path: A model path cannot be reproduced by real traces, indicating a discrepancy.
  • Certification Process Adaptation: Based on the identified inconsistencies, the accredited lab adapts the certification process. This can range from no adaptation (for minor issues) to re-execution of tests, partial re-certification, or a full re-certification if critical inconsistencies are found. This adaptive approach aims to maintain certificate validity without requiring costly full re-certifications for every minor change.

Datasets

Not discussed in the paper. The authors mention generating an experimental dataset for evaluation, but do not specify publicly available datasets.

Key Insights and Ideas

  • Runtime Verification is Crucial: The dynamic nature of the cloud necessitates continuous verification of service behavior after initial certification. Relying solely on offline testing is insufficient.
  • Model as the Ground Truth: A formal model (FSM) serves as the benchmark against which actual service behavior is continuously compared. The trustworthiness of the certification hinges on the accuracy of this model.
  • Inconsistency-Driven Adaptation: Identifying specific types of inconsistencies (partial path, new path, broken path) provides actionable insights for adapting the certification process, making it more efficient and less disruptive.
  • Bridging the Gap: The approach aims to bridge the gap between the theoretical model used for certification and the actual, potentially evolving, implementation of cloud services.

Key Results

The paper presents an experimental evaluation of a prototype implementation. While specific quantitative results are not detailed in terms of accuracy metrics like precision or recall, the evaluation demonstrates the effectiveness of their model adapter. The refined models achieved an average coverage of 64% of the original correct model paths, representing a 28% increase in coverage compared to the inconsistent models. Notably, 17% of the refined models achieved 100% coverage of the original model. The study also highlights that as the perturbation (inconsistency) level increased, the ability to achieve full coverage decreased, but even with significant perturbation, a substantial portion of the original model was recovered.

Limitations and Future Work

The authors acknowledge a limitation: their current model verification approach does not explicitly model cloud infrastructure configurations. This can lead to scenarios where a property guaranteed in a lab environment might not hold in a production environment due to differences in shared resources or other infrastructure-specific factors. They suggest this as a direction for future work.

Additionally, they propose applying fuzzing and mutation techniques to inconsistent models as a way to reveal hidden paths that might not be discovered through current trace collection methods. This would further enhance the robustness of their model verification and adaptation process.

Summary updated

BibTeX
@inproceedings{anisetti2014trustworthy,
  author = {Marco Anisetti and Claudio A. Ardagna and Ernesto Damiani and Nabil El Ioini},
  title = {{Trustworthy Cloud Certification: A Model-Based Approach}},
  booktitle = {Simpda (revised Selected Papers)},
  year = {2014},
  doi = {10.1007/978-3-319-27243-6\_5},
}
Figure