On Global Fragmentation Metrics as Proxy for Network Blocking: Correlation, Detection and Prediction

Elastic Optical Networks (EONs) are challenged by spectrum fragmentation, which can obstruct the establishment of new connections. While the concept of fragmentation gaps is relatively simple, determining an accurate metric to quantify fragmentation remains complex, and numerous metrics have been proposed to address this. In this work, we analyze the effectiveness of various fragmentation metrics in understanding and forecasting network blocking. We begin by evaluating the correlation between common global fragmentation metrics and blocking probability, finding a relationship between the two. Through the application of a Random Forest model, we identify the most useful metrics for anticipating blocking caused by fragmentation. Our classification model further assesses whether these metrics can accurately detect current blocking and predict future blocking events. The findings reveal that although global fragmentation metrics perform better than chance, they are not yet highly reliable predictors, opening avenues for future exploration of more detailed fragmentation metrics.