Softwarized Edge Intelligence for Advanced IIoT Ecosystems: A Data-Driven Architecture Across the Cloud/Edge Continuum

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article proposes a soft defined edge intelligence architecture for the Industrial Internet of Things (IIoT), which achieves cloud edge end collaborative data processing and low latency response through containerized microservices, Kubernetes orchestration, and ML driven dynamic reconstruction mechanism. The research question has practical significance, the architecture design is complete, and the experimental verification is detailed, making a positive contribution to promoting the intelligence of IIoT systems. But some technical details need further clarification, and the limitations of the experiment need to be discussed and supplemented.

1. Discuss the gap between simulation environments (such as Mininet link models) and industrial field networks (time sensitive networks TSN).
2. Increase model selection experiments (such as computational cost/accuracy trade-offs) to demonstrate the feasibility of SVM in edge resources.
3. Discuss the applicability of 800ms reconstruction delay for typical IIoT applications (predictive maintenance vs. real-time control).
4. Supplement the terminology list (such as the Abbreviations list).
5. Add comparative baselines in the experimental results to highlight the advantages of this architecture.

This article demonstrates solid performance in architecture design and experimental verification, addressing key challenges such as cloud edge collaboration and dynamic reconstruction in IIoT. However, there is a need to strengthen industrial scenario verification and comparative analysis.

Author Response

Please see the attachment. Thank you.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this work, the authors proposed an architecture to operate Edge AI models within the IIoT continuum. The authors claim this architecture is data driven and softwarized that integrate cloud, edge, and IoT devices. The proposed architecture, exemplified in Figure 1, displays the hierarchy of the proposed application.

On the one hand, the most positive aspect of this work is its completeness. The authors presented several steps to run the system according to its proposal. From a software engineering perspective, the proposal seems very much ready to launch in a more mature environment, or even in a real industrial plants. On the other hand, the running on simulated environments hide some important decisions, which are essential to edge computing.

For instance, the authors cite several networking protocols as a baseline for the edge-to-cloud communication. Nevertheless, some of these protocols are still theoretical, such as 6G, and others need some guarantees to run in industrial environments. Depending on the industrial activity, 5G is not available, nor WiMAX, Wi-Fi, etc. It is actually common to lack support on reliable wireless networking. Is the architecture reliable to Industrial Environments without adequate communication bandwidth?

Another important question raises from the simulated environment is the behavior of federated learning and embedded machine learning. The authors chose to use an SVM as model to classify the data. Why? Even if it presented adequate scores, it can underperform in embedded devices both in inference speed and scoring. Why not using a small DNN, for instance? If the model is going to be embedded or retrained on edge devices, it is essential to understand the deployment protocol for this model, its memory footprint, etc.
 
 Finally, there were some timing aspects discussed, but the work lacks on assessing real-time or QoS (which are similar in the IoT context). The authors should carefully asses in their text what belongs to their context and what does not.

Even with these issues, the text has a high contribution, and is very mature. This reviewer supports the publication of this work after the authors assess these aspects.

Author Response

Please see the attachment. Thank you.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The draft titled “Softwarized Edge Intelligence for Advanced IIoT Ecosystems: A Data-Driven Architecture Across the Cloud/Edge Continuum” addresses an interesting point, The work has the potential to make significant contribution to the field.

In this paper, the authors put forward and assess a Kubernetes-based, microservice-oriented architecture that seamlessly integrates cloud, edge, and factory-floor layers. The manuscript is well written  and the experimental evaluation showcases its practical feasibility and scalability. The inclusion of open-source code and deployment scripts really enhances its value for reproducibility. Overall, the writing is clear, and the related work section provides a thorough overview of IIoT architectures. The integration of machine learning inference, dynamic network reconfiguration, and monitoring tools like InfluxDB/Grafana really bolsters the paper’s contribution.

While the related work section is quite comprehensive, it would be beneficial to better contextualize the experimental results against existing architectures (like SEGA or blockchain-based IIoT).

A quantitative or qualitative comparison  is not clear. A quantitative or qualitative comparison could really highlight the unique advantages of the proposed approach. The manuscript emphasizes performance, scalability, and orchestration, but it only briefly addresses security.

Discussion on confidentiality, authentication, and resilience to cyberattacks is missing. Given the critical nature of IIoT, a discussion on how the proposed architecture ensures data confidentiality, authentication, and resilience to cyberattacks would significantly enhance its contribution.

The evaluation relies on a Kaggle dataset to simulate IIoT sensors. While this is acceptable for proof-of-concept, the paper should more clearly outline the limitations of using synthetic or emulated data compared to real industrial deployments and suggest how future work might tackle this.

It is suggested to expanding the discussion to include additional scenarios like smart energy grids, logistics, or healthcare  would broaden the applicability and impact of the proposed framework.

The paper presents active-standby mechanisms for stateful components, but more emphasis on failure scenarios (e.g., controller crash, communication loss) and quantitative evaluation of recovery time would be valuable.

Some diagrams are quite dense. Simplifying them or adding explanatory captions would Improve readability. Similarly, Algorithm 1 could be better structured for clarity, perhaps using pseudocode conventions.

 

 

Author Response

Please see the attachment. Thank you.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript presents an architecture for Industrial Internet of Things systems, leveraging edge computing and machine learning for real-time decision-making and network reconfiguration. While the work shows promise in addressing key challenges in IIoT environments, several critical aspects need further refinement and validation.

Firstly, the study relies heavily on a simulated environment using Mininet-WiFi for the factory floor, which may not fully capture the complexities and variability of real-world industrial settings. The authors should consider integrating real IIoT hardware into their testbed to validate the architecture under more realistic conditions. This would provide a more accurate assessment of the system's performance and robustness in practical applications. Additionally, the dataset used for training the machine learning model is based on a publicly available source, which may not be representative of the diverse and dynamic data generated in actual industrial environments. The authors should explore the use of more comprehensive and varied datasets to ensure the model's generalizability and effectiveness across different IIoT scenarios.

Secondly, while the proposed architecture demonstrates scalability and efficient resource utilization, the evaluation lacks a detailed comparison with existing state-of-the-art solutions. The authors should conduct a thorough benchmarking analysis, comparing their architecture with other advanced IIoT frameworks in terms of latency, throughput, and resource consumption. This would provide a clearer understanding of the proposed system's advantages and limitations in the context of current research advancements. Furthermore, the blocking probability analysis under different QoS thresholds is informative, but it would be beneficial to include additional metrics such as system reliability, fault tolerance, and recovery time to provide a more comprehensive evaluation of the architecture's performance in dynamic industrial environments.

Author Response

Please see the attachment. Thank you.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Well done.

Author Response

Thank you very much for all your comments during the different steps in the review.