SDN Fabric Saturation Prediction via ML Flow Admission

The patent applies preliminary action by implementing saturation prediction models that forecast future fabric saturation conditions before they occur. The system analyzes historical traffic patterns and flow characteristics to predict when the SDN fabric will become saturated, allowing proactive flow admission decisions to be made in advance, thereby improving resource utilization without requiring complex real-time control mechanisms.

The patent implements feedback mechanisms where the saturation prediction model continuously monitors actual fabric utilization and compares it with predicted values. Machine learning algorithms adjust flow admission control parameters based on this feedback, creating a closed-loop system that optimizes resource utilization dynamically while adapting to changing traffic patterns, thus managing complexity through adaptive rather than static control.

The system performs preliminary computation by training saturation prediction models offline using historical traffic data. Once trained, these models can quickly predict saturation conditions for new flows without requiring complex real-time computation, thus achieving high prediction accuracy while minimizing processing time during actual flow admission decisions.

The patent applies partial action by using simplified machine learning models that capture the most critical patterns in traffic data without requiring exhaustive analysis. The system focuses on key features and patterns that are most predictive of saturation, rather than analyzing all possible variables, thereby achieving sufficient prediction accuracy with reduced computational overhead and processing time.