Using Machine Learning to Provide Reliable Differentiated Services for IoT in SDN-Like Publish/Subscribe Middleware.

Area of Science:

• Computer Science
• Networking
• Internet of Things

Background:

• Existing publish/subscribe middleware assumes uniform Quality of Service (QoS) requirements for all users.
• Internet of Things (IoT) scenarios often involve diverse user needs, particularly regarding differing delay tolerances.
• Providing reliable differentiated services in such environments is a significant challenge.

Purpose of the Study:

• To propose and evaluate an SDN-like middleware architecture for differentiated QoS in IoT environments.
• To leverage Software-Defined Networking (SDN) programmability for customized service delivery.
• To enhance the reliability and performance of publish/subscribe systems by addressing varying user requirements.

Main Methods:

• Development of an SDN-like publish/subscribe middleware architecture.
• Integration of priority queues within OpenFlow switches for differentiated service handling.
• Application of the eXtreme Gradient Boosting (XGBoost) machine learning model to accurately predict switch queuing delay.
• Implementation of a two-layer queue management mechanism for reliable differentiated service guarantees.

Main Results:

• The XGBoost model demonstrated high accuracy in predicting switch queuing delays compared to real-world values.
• The proposed two-layer queue management mechanism effectively reduced end-to-end delay.
• The system achieved a lower packet loss rate and more reasonable bandwidth allocation.
• Experimental evaluations validated the effectiveness of the SDN-based approach for differentiated QoS.

Conclusions:

• The proposed SDN-like middleware architecture successfully enables differentiated services in IoT publish/subscribe systems.
• The combination of SDN programmability, accurate delay prediction, and a two-layer queue management mechanism significantly improves QoS.
• This approach offers a viable solution for meeting diverse user delay requirements in real-world IoT applications.