IP networks monitoring with BigData - A passive approach for large-scale networks’ performance analysis

Monitoring of traffic in an IP network is a crucial point to prevent big failures of the system and to deliver a good quality of service to end-users. This is a key factor of success for large-scale Internet Service Providers. For this reason, Telecom Italia developed a series of techniques that have been defined as Alternate Marking Performance Monitoring. The main requirements of this system are scalability, robustness, and flexibility. This work aims to develop and test a system able to perform passive network monitoring exploiting Alternate Marking techniques in a multipoint environment. Starting from several works that have been carried out in the past, critical points that have been highlighted has been addressed, proposing a solution for some of them. In order to powerfully monitor performances, it is necessary to split the network into balanced parts, so that the same level of detail can be detected in each of them. To tackle this problem, it has been necessary to formalize it, and then propose a possible solution that exploits Deep Reinforcement Learning. The importance of performance metrics is strictly related to the time needed for producing them: data that have been generated almost in real-time can help to anticipate problems. In this work, a possible architecture that is robust and deployable in a real environment has been proposed, also providing the possibility to produce data in a real-time manner. The proposed architecture has been emulated on a large-scale network, obtaining metrics that have been compared with expected performances. Thanks to these results, the viability of the implemented solution has been shown. Finally, possible future works have been highlighted in the direction followed by this thesis in order to close the gap between a research and development environment towards a real deployed network.