ARMA-BASED TRAFFIC PREDICTION AND OVERLOAD DETECTION OF NETWORK

With its rapid development the network now has a large size and complexity, and consequently the network management is becoming increasingly difficult. Generally the network manager begins to solve the potential problems after the monitoring system alarms, i.e., it takes a re action way, so the service on the network is possibly affected. A normal behavior model is founded from the numbers of the non unicasting packet collected from a real network. The normal behavior serial is stabilized and the coefficients of the ARMA model are estimated. And then the traffic by the way of "minimal linear square error" is predicted, and the probability of the predicted value exceeding the threshold is calculated. So the overload in the network may be predicted, and the recovery measures may be taken beforehand to prevent communication from being impacted or to reduce its severity. This method changes the traditional network management from re action to prediction beforehand, so that the network overload may be predicted.