Traffic Forecasting for Monitoring in Computer Networks using Time Series

With the expansion of connectivity and information exchange, monitoring Internet traffic becomes a priority in network management to identify anomalies and resource use. This paper presents a study of data traffic forecasting on a computer network, by using known approaching methods for Time Series analysis. The objective of this work is to monitoring the connection of users to network-based applications, including resource availability and network stability of a Brazilian educational institute. To estimate the traffic at a given time, the adjustments made with Exponential Smoothing, AR and ARIMA models were compared in order to detect possible future abnormal behavior of network usage. The results indicate that the chosen models, mainly the ARIMA, can be used to predict both input and output traffic of a network, also allowing the generation of alerts in real time. It is possible to predict how Internet traffic will be in the next few moments in order to detect possible anomaly on the network in a short period of time when they differ considerably from the forecast made for that specific period. Efficient network monitoring favors the quality of applications and services available to users, helping the network manager to make decisions for maintenance and constant improvement.