Virtualized 4G long term evolution testbed to investigate the performance impact of in-band network telemetry

Abstract

As the fifth generation of mobile networks move into the cloud, there is a need to develop new and more flexible monitoring systems. In-band Network Telemetry (INT) has begun to make its mark on the Network Monitoring field, and provides a new way of collecting and processing network telemetry data directly on programmable switches. With INT we can use the in-band network traffic itself as the carrier for monitoring metrics by leveraging the data plane of programmable switches. In this thesis we develop a fully virtualized 4G Long Term Evolution (LTE) testbed in a VirtualBox environment with INT support, and we develop an algorithm that leverages the TOS field in IP Packet Headers to detect Packet Loss and Network Delay between two P4-capable BMV2 switches. The Radio Access Network (RAN) and and Evolved Packet Core (EPC) components of the LTE network are implemented with the OpenAirInterface project, and the EPC components are deployed in a Mininet environment as containers, which are connected to P4-capable Behavioral Model V2 (BMV2) switches. INT does incur a higher cost to CPU compared to more traditional monitoring systems such as passive and active systems, but maintains a higher measured bandwidth even with this cost. INT is also able to detect network delay with high reliability, but has reduced accuracy for detecting packet loss.