Evaluating predictors of congestion collapse in communication networks

Abstract

Congestion in communication networks can be modeled as a percolation process, where congestion spreads minimally before a critical load and expands rapidly afterwards. Some studies identify predict onset of rapidly expanding congestion in time to alert network managers to take mitigating actions to avoid congestion collapse. The paper specifies five predictors: autocorrelation, variance, threshold, growth persistence, and growth rate. Predictor performance is measured for three simulated network models, under two traffic scenarios: increasing and steady load. Predictors are compared on implementation cost, accuracy, warning time, and persistence. The rates and types of prediction errors are also characterized. Results showed that: (1) predictor performance is influenced by network-model realism; (2) the autocorrelation and variance predictors performed poorly in some situations; (3) the threshold predictor yielded best overall accuracy, with mean warning time exceeding seven minutes for the most realistic network model. The paper also suggests a necessary condition to control false positives.