Evaluation of FPGA-based In-band Telemetry Methodologies for the Responsible Internet

As digital infrastructures become increasingly critical to society, concerns over their trustworthiness and reliability have gained attention among European policymakers. The concept of Responsible Internet emerges as the next pivotal phase in their evolution, underlining the necessity of transparency in networks granting both critical service providers and everyday users the ability to know where the data flows they inject into the network are routed and which chain of providers handles them. To achieve such a goal and contribute to dismantling the black box nature of networks, we explored In-band Network Telemetry, a framework for collecting fine-grained metadata about the network's state with low and fixed transmission overhead. This technique relies on programmable data planes supported by modern yet uncommon network devices, such as the FPGA-based SmartNICs. It modifies packet processing by embedding telemetry operations into the forwarding plane, thus enabling telemetry data collection without the need for network sampling. In this thesis, our focus was on developing data plane telemetry solutions utilizing the P4 programming language, which allows for the flexible programming of network devices, such as FPGAs, to enable custom packet processing and real-time telemetry data collection directly in the forwarding plane. Experimental validation was conducted within the FABRIC International Testbed, which enables research at scale across scientific domains. This work contributes to bridging the gap between network transparency and practical deployment, leveraging cutting-edge testbed environments and paving the way for more transparent and accountable infrastructures.