Time series data mining for network service dependency analysis

Abstract

In data-communication networks, network reliability is of great concern to both network operators and customers. To provide network reliability it is fundamentally important to know the ongoing tasks in a network. A particular task may depend on multiple network services, spanning many network devices. Unfortunately, dependency details are often not documented and are difficult to discover by relying on human expert knowledge. In monitored networks huge amounts of data are available and by applying data mining techniques, we are able to extract information of ongoing network activities. Hence, we aim to automatically learn network dependencies by analyzing network traffic and derive ongoing tasks in data-communication networks. To automatically learn network dependencies, we propose a methodology based on the normalized form of cross correlation, which is a well-established methodology for detecting similar signals in feature matching applications.