Towards the improvement of performance anomaly prediction

Growing demand for pro-active abilities in network management requires performance monitoring agents not only to be able to monitor the anomalies, but also to predict future occurrences. Recent research in this area would usually apply a neural network algorithm on raw SNMP or NetFlow data to obtain the knowledge about the patterns in performance data. The results are not always satisfactory due to highly unpredictable nature of cross-traffic in the network. This paper attempts to improve the prediction quality by using data obtained from end-to-end probing. The results prove higher resilience to cross-traffic interference and better pattern recognition.