TemporalEMF: A Temporal MetamodelingFramework - Extended Abstract

Abel Gómez, Jordi Cabot, Manuel WimmerINFORSID 2019 · OpenReview

Abstract

In Model-Driven Engineering (MDE), models typically represent static snapshots of a system's state. This limitation hinders the ability to analyze system evolution, perform historical audits, or debug past states without manual and labor-intensive retrieval from version control systems. This paper introduces TemporalEMF, a framework designed to integrate temporal awareness directly into the Eclipse Modeling Framework (EMF). By leveraging a NoSQL backend (Apache HBase) and extending the Object Constraint Language (OCL), TemporalEMF enables efficient storage, retrieval, and querying of model history while maintaining full compatibility with existing modeling tools.

Introduction

Model-Driven Engineering (MDE) treats models as the primary artifacts of the software development lifecycle. However, the vast majority of current modeling environments are "temporally blind," meaning they only capture the current state of a model. When engineers need to understand how a system configuration evolved or identify the state of a model prior to a failure, they are often forced to rely on external versioning systems like Git. These systems are not optimized for model-level granularity, making complex temporal queries—such as "find all versions where this component's status was 'active'"—nearly impossible to execute efficiently.

TemporalEMF addresses this gap by transforming the EMF ecosystem into a temporally-aware environment. The goal is to make model history a first-class citizen without requiring developers to abandon their existing tools or learn entirely new modeling paradigms.

The TemporalEMF Framework

The architecture of TemporalEMF is built upon three core pillars designed to balance functionality with ease of adoption:

1. Lightweight Metamodel Extension

Rather than forcing a complete overhaul of existing metamodels, TemporalEMF utilizes a "profile" approach. This allows developers to annotate existing metamodels with temporal metadata. By extending the metamodel rather than replacing it, the framework ensures that the structural integrity of the original design remains intact while gaining the ability to track changes over time.

2. NoSQL Infrastructure (Apache HBase)

To manage the high volume of data generated by tracking every atomic change in a model, TemporalEMF utilizes Apache HBase. HBase is a distributed, versioned NoSQL database modeled after Google’s BigTable. It is uniquely suited for temporal modeling because it natively stores data in a (row, column, timestamp) format. TemporalEMF maps model elements and their attributes directly to this structure, allowing the database to handle versioning at the cell level. This architectural choice minimizes the overhead typically associated with managing historical deltas manually.

3. Temporal Query Language

To enable sophisticated analysis of model evolution, the authors extended the standard Object Constraint Language (OCL). OCL is widely used in the EMF ecosystem for defining constraints and querying model states. The extended temporal OCL introduces operators that allow users to query the model across the time dimension, enabling developers to write expressions that analyze how specific model elements have changed or interacted over a defined period.

Evaluation: Industry 4.0 Case Study

The framework was evaluated using an Industry 4.0 case study involving a production system simulator. In this scenario, the simulator generated large-scale, rapidly evolving models representing the state of a factory floor.

The evaluation focused on two primary metrics:

  • Scalability: The framework successfully handled high-frequency updates from the simulator, demonstrating that the HBase backend can scale with the data demands of real-time industrial simulations.
  • Transparency: Because TemporalEMF acts as a transparent "add-on," the existing simulator logic did not require modification. Tools that were not designed for temporal awareness simply interacted with the "latest" version of the model, while temporal-aware tools could dive into the history via the same interface.

Key Insights

  • The "Add-on" Philosophy: By maintaining compatibility with the standard EMF API, TemporalEMF avoids the tool fragmentation that often plagues new MDE research. It allows a decade of existing modeling tools to benefit from temporal features without being rewritten.
  • Model Evolution as Data: TemporalEMF shifts the perspective of a model from a static file to a continuous stream of states. This perspective is critical for "model mining," where researchers analyze design patterns and evolution to predict future errors or optimize system performance.

Conclusion and Future Work

TemporalEMF provides a robust foundation for temporal modeling, but several avenues for future research remain. The authors plan to expand the framework's flexibility by supporting additional NoSQL backends beyond HBase. Furthermore, there is a focus on improving the user experience by integrating temporal querying into web-based modeling environments.

Looking forward, the ultimate goal is to move from reactive historical analysis to predictive modeling. By using the accumulated historical data stored in TemporalEMF, the authors aim to apply machine learning techniques to classify user behavior and predict future modeling states, potentially identifying design flaws before they are even committed.

Summary updated

BibTeX
@inproceedings{gomez2019temporalemf,
  author = {Abel Gómez and Jordi Cabot and Manuel Wimmer},
  title = {{TemporalEMF: A Temporal MetamodelingFramework - Extended Abstract}},
  booktitle = {INFORSID},
  year = {2019},
  url = {https://openreview.net/forum?id=9jxrGiAIzB},
}
Figure