A Model-Centric DevOps Architecture for DEVS-Based Digital Twin Simulation Services

A Model-Centric DevOps Architecture addresses the lack of versioning and continuous delivery in DEVS-based digital twin simulations by treating simulation models as first-class DevOps artifacts. The architecture introduces a Simulation Model Definition Language (SMDL), a declarative YAML-based language with a formal mapping to the multiPDEVS formalism, enabling automated structural and semantic validation within a CI/CD pipeline.

Key features include: Automated Validation: Structural checks (port resolution, coupling closure) and semantic checks run in CI/CD without invoking the simulation runtime, ensuring model correctness before deployment. Immutable Versioning: Models are stored as versioned immutable artifacts in object storage, allowing for easy rollback by pinning a specific version identifier. Cloud-Native Deployment: The simulation engine is adapted into containerized microservices on Kubernetes, supporting elastic scaling and isolated execution of simulation scenarios. Reproducible Workflows: The approach transforms ad hoc model management into automated, reproducible workflows, enabling domain experts to iterate on model configurations (e.g., network parameters) without modifying engine code.

While the architecture provides a robust framework for integrating DEVS simulations into modern DevOps practices, recent reviews note that empirical evidence demonstrating the pipeline's ability to catch malformed models (e.g., error-injection studies) is currently deferred to companion studies.

The paper addresses a lifecycle-management gap in DEVS-based digital twin simulation: simulation models are frequently revised, reconfigured, and redeployed in ways that resemble software evolution, but many DEVS simulation environments still lack practical support for model versioning, automated validation, and continuous delivery. This is especially problematic in cloud-native settings, where digital twin services must be scaled, monitored, updated, and rolled back with the same operational discipline applied to microservices. The work frames the issue as a mismatch between the formal strengths of DEVS—compositional semantics, well-defined event interaction, and rigorous simulation structure—and the engineering practices now expected for production simulation services.

Its central contribution is a model-centric DevOps architecture for DEVS-based digital twin simulation services. Rather than treating the simulation engine as the primary deployable unit, the architecture elevates the DEVS model itself to a first-class artifact with explicit versioning, dependency metadata, validation gates, and deployment descriptors. This enables simulation models to be built, tested, packaged, and promoted through continuous-integration/continuous-delivery pipelines, while still preserving the formal semantics of the underlying DEVS representation. In effect, the paper proposes a bridge between formal discrete-event modeling and modern cloud-native service operations, making model evolution more reproducible, auditable, and operationally manageable.

The work matters because it targets one of the practical bottlenecks in deploying digital twins at scale: the ability to update simulation logic continuously without sacrificing correctness, traceability, or operational stability. By aligning DEVS model management with DevOps practices, the architecture can reduce manual redeployment effort, support regression-style validation of model changes, and enable safer rollout of new simulation behavior in production. For technically oriented readers, the contribution is less about a new simulation algorithm and more about a systems-level design pattern that makes formal simulation models viable as continuously delivered cloud services.