Detection and localization of events in computer networks using end-to-end measurements time series

Abstract

To better understand the performance of its own network, a major tier-3 Brazilian ISP, in partnership with UFRJ and a startup incubated at this university, established a project to monitor the service provided to a subset of its customers. In this context, guided by the specific ISP’s network’s characteristics, and the current measurement process, this dissertation aims to check the viability of only using end-to-end QoS measures, and traceroutes, to identify and localize network events. An event can be interpreted as a behavioral change in a network equipment, that affect the quality of service perceived by the end-users, such as a router failure. The localization procedure defines a set of feasible locations where the event could have happened. For such purpose, this work proposes a data analytics framework, which is able to track statistical changes in the QoS time series of different clients. To detect and localize events, the mechanism correlates these modification patterns with traceroutes. In order to increase the system’s performance, this dissertation also indicates possible improvements in the current measurement methodology.