PFA-INT: Lightweight In-Band Network Telemetry with Per-Flow Aggregation

In-band network telemetry (INT), empowered by programmable dataplanes such as P4, comprises a viable approach to network monitoring and telemetry analysis. However, P4's INT and other existing frameworks for INT yield a substantial transmission overhead, which grows linearly with the number of hops, as well as with the number of telemetry variables.In this respect, we present a lightweight INT framework that exercises per-flow aggregation (PFA), i.e., spreads the telemetry values across the packets of a flow. In order to enable the required switch coordination, without any centralized controller intervention, we utilize per-flow telemetry states, which are maintained within P4 switches. We further leverage on Bloom Filters in order to compress the state lookup tables within P4 switches. Our proposed INT framework, namely PFA-INT, encompasses all required functionality for the continuous tracing of telemetry values over the entire duration of flows. Our evaluation results indicate that PFA-INT yields high monitoring accuracy with a fixed transmission overhead. We also study potential implications on path tracing, when the BF collision rate is increased. Even in such conditions, the continuous tracing of monitoring values help alleviate such implications and correctly retrieve telemetry values.