Abstract :
[en] Modern networks are becoming increasingly complex, driven by high-speed links, virtualization, encrypted transport protocols, and the proliferation of globally distributed
services. These networks rely on large-scale content distribution infrastructures, making
their behavior highly dynamic and often difficult to predict. Operators must ensure both high
availability and high-quality user experiences, yet traditional monitoring and measurement
approaches struggle to provide timely, scalable, and fine-grained visibility. Active probing can be intrusive and limited in coverage, while passive collection often lacks sufficient detail to capture per-hop performance and path characteristics. This growing complexity and the limitations of conventional tools highlight the urgent need for advanced observability mechanisms that operate within the data plane, capture precise telemetry in real time, and remain deployable across production environments. This thesis addresses these challenges by exploring the use of IPv6 Extension Headers and In Situ Operations, Administration, and Maintenance (IOAM) to enable robust observability. In IPv6, Extension Headers allow optional metadata to be carried in packets without modifying the core protocol, but their behavior in operational networks is inconsistent: routers and middleboxes may process or ignore Extension Headers, or worse, drop IPv6 packets containing Extension Headers. As for IOAM, which uses Extension Headers when used with IPv6, it allows packets to record per-hop operational data, capturing detailed latency, path, and network events in general. Beyond
evaluating these mechanisms individually, this work investigates their combination to enable advanced observability, correlating network telemetry with application-level metrics for a more comprehensive, context-aware understanding of system performance. The main contributions of this thesis are: (i) a thorough evaluation of the survivability and processing behavior of IPv6 Extension Headers in controlled environments and at Internet scale; (ii) the implementation of IOAM in the Linux kernel, including performance evaluation; and (iii) the development and evaluation of a new technique that integrates network and application telemetry to provide richer operational insights. Together, these contributions advance the state of the art in observability, demonstrating how existing protocols and mechanisms can be leveraged to meet the evolving needs of modern networks while enabling operators to better understand, manage, and optimize complex infrastructures.
