Towards probe planning for in-band network telemetry

Abstract

In-Band Network Telemetry (INT) has emerged as a powerful network monitoring approach in programmable networks, providing fine-grained visibility into network events. However, existing INT orchestration approaches often overlook fault tolerance in the data plane, leaving monitoring mechanisms compromised during network failures. To address this gap, we propose InPatching, an in-network fault-tolerant approach for INT-based monitoring. InPatching autonomously detects faulty devices and applies coordinated detours in affected probing cycles, ensuring uninterrupted telemetry data collection without relying on the control plane. By offloading recovery to the data plane, InPatching significantly reduces the recovery time compared to control plane strategies. To enable efficient detours, we formalize fault-tolerant probing planning for INT using a Mixed-Integer Linear Programming (MILP) model. This model allows us to efficiently determine the optimal detour paths and minimize the impact on network performance. Our extensive evaluation demonstrates the effectiveness of InPatching in comparison to control plane solutions. We show that InPatching outperforms control plane approaches by a factor of 18X, providing fast and reliable recovery for INT-based monitoring while avoiding substantial overhead. The compilation of the code into hardware has also been successfully performed, and the metrics obtained regarding resource usage – i.e., MAU and Tagalong collections –indicate low memory resource utilization, on average, for allocating the rewritten code components in the TNA architecture. In addition to the technical contributions, we also release open-source software artifacts that facilitate the adoption and reproducibility of InPatching. Network operators can leverage this solution to maintain network-wide visibility even during network failures, ensuring continuous coverage and freshness of collected INT data. Overall, our work contributes to the advancement of fault-tolerant network monitoring and highlights the importance of considering data plane resilience in the design of INT orchestration approaches. By addressing this critical aspect, InPatching significantly enhances the reliability and effectiveness of INT-based monitoring systems in programmable networks. Key-words: In-band Network Telemetry (INT), Software-Defined Network(SDN), Probe, Network Monitoring, Fast Reroute (FRR)