A predictability analysis of network traffic

Abstract

This paper assesses the predictability of network traffic by considering two metrics: (1) how far into the future a traffic rate process can be predicted for a given error constraint; (2) what the minimum prediction error is over a specified prediction time interval. The assessment is based on two stationary traffic models: the auto-regressive moving average (ARMA) model and the Markov-modulated Poisson process (MMPP) model. Our study in this paper provides an upper bound for the optimal performance of online traffic prediction. The analysis reveals that the application of traffic prediction is limited by the quickly deteriorating prediction accuracy with increasing prediction interval. Furthermore, we show that different traffic properties play different roles in predictability. Traffic smoothing (low-pass filtering) and statistical multiplexing also improves predictability. In particular, experimental results suggest that traffic prediction works better for backbone network traffic, or when short-term traffic variations have been properly filtered out. Moreover, this paper illustrates the various factors affecting the effectiveness of traffic prediction in network control. These factors include the traffic characteristics, the traffic measurement intervals, the network control time-scale, and the utilization target of network resources. Considering all of the factors, we present guidelines for utilizing and evaluating traffic prediction in network control areas.