A Modeling Approach to Support the Similarity-Based Reuse of Configuration Data

Razieh Behjati, Tao Yue, Lionel C. BriandMoDELS 2012 · DOI

TL;DR

This paper introduces a model-based approach to automatically reuse configuration data in complex embedded systems (Integrated Control Systems or ICSs). By modeling internal similarities within a product family, the authors enable significant reduction in manual configuration effort, saving over 60% of decisions in their industrial case study.

What Problem Does This Paper Address?

Modern embedded systems, particularly Integrated Control Systems (ICSs) like industrial robots and process plants, are becoming increasingly complex. These systems are often built from a generic software base that requires extensive configuration for each specific product. This configuration process involves manually setting tens of thousands of parameters, making it time-consuming, error-prone, and challenging. While existing approaches focus on high-level variability in feature models, they often fall short in capturing the intricate architectural variabilities and dependencies found in highly configurable embedded systems. The core problem this paper addresses is the lack of automated support for reusing configuration data based on the inherent structural similarities within a product family, which leads to significant manual effort and potential for errors.

Methods

The authors propose a four-step, model-based approach to automate configuration reuse:

  1. Product-Family Modeling: This step uses the SimPL methodology to create a generic, configurable model of an ICS family. This model captures the architecture, variabilities, and consistency rules. This step is standard and builds on prior work.
  2. Similarity Modeling: This is the core novelty. Domain experts manually create "similarity models" that express structural similarities within a product. These models operate at two levels:
  3. Low-level (OCL): Object Constraint Language (OCL) is used to define specific similarity rules as constraints on model elements. These rules specify that certain configurable parameters must have identical values across different instances of components within a product.
  4. High-level (Feature Models): Feature models are used to provide a user-friendly representation of these similarity rules and their variabilities. This allows configuration engineers to select which similarity rules apply to a given product without needing to understand OCL.
  5. Similarity Configuration: Configuration engineers use the similarity feature models to select applicable similarity rules for a specific product. This selection results in a "similarity specification" (a collection of OCL constraints). This step gives engineers control over the reuse process.
  6. Guided Configuration: The generic SimPL model and the product-specific similarity specification are fed into an automated configuration engine. This engine uses a constraint solver to infer values for configurable parameters based on the selected similarity rules. When a parameter's value is set, the similarity rules automatically propagate that value to related components, enforcing reuse and ensuring consistency. This step extends their previous work by incorporating the similarity specification.

Datasets

The paper evaluates its approach using two complete subsea products from an industrial partner. These products are described as representative of their domain, with significant internal dissimilarities between them, making them a challenging test case. The generic software model for this case study contains 36 configuration units and 264 configurable features. The similarity feature model derived from these products has a depth of four, with 200 features, including 81 leaf features representing similarity rules.

Key Insights and Ideas

  • Internal vs. Common Similarities: The paper distinguishes between "commonalities" (similarities across different products in a family) and "internal similarities" (similarities among different parts within a single product). The focus here is on leveraging internal similarities for reuse.
  • User Control via Feature Models: Using feature models for similarity rules is crucial. It abstracts away the complexity of OCL, allowing configuration engineers to easily select and manage which similarities should be applied, thereby controlling the degree of automated reuse.
  • Automated Inference through Constraint Propagation: The core mechanism for reuse is the constraint solver. By incorporating similarity rules (expressed as equality constraints) into the system, setting one parameter automatically infers values for others, leading to significant automation.
  • Refactoring for Granularity: The paper discusses refactoring coarse-grained similarity rules into finer-grained ones. This allows for more flexible selection of similarities and better management of dependencies, especially when requirements evolve.
  • Practical Significance: The approach is demonstrated on a real-world industrial case study, highlighting its applicability and substantial benefits for large-scale, highly-configurable systems.

Key Results

The evaluation on two industrial subsea products showed significant reuse rates:

  • Product 1: 52 similarity rules, 263 equality relations, resulting in 19,289 automated decisions and a reuse rate of 0.647 (64.7% of configuration decisions automated).
  • Product 2: 41 similarity rules, 270 equality relations, resulting in 46,801 automated decisions and a reuse rate of 0.834 (83.4% of configuration decisions automated).

The authors state that this automation can save more than 60% of the configuration effort compared to a scenario with no reuse support. This is achieved by automatically inferring a large portion of the configuration decisions.

Limitations and Future Work

  • Manual Effort for Similarity Modeling: The authors acknowledge that creating similarity models is manual and time-consuming. However, they argue that this upfront effort is paid back by the significant reuse achieved across all products in a family.
  • Generalizability: While the approach is demonstrated on ICSs with highly-symmetric hardware, the authors expect it to generalize to other domains with similar embedded software or distributed networked systems.
  • Future Work: The authors plan to conduct experiments with human subjects to further evaluate the applicability of their approach.

Summary updated

BibTeX
@inproceedings{behjati2012modeling,
  author = {Razieh Behjati and Tao Yue and Lionel C. Briand},
  title = {{A Modeling Approach to Support the Similarity-Based Reuse of Configuration Data}},
  booktitle = {MoDELS},
  year = {2012},
  doi = {10.1007/978-3-642-33666-9\_32},
}
Figure