IRIS: An Interactive Programmable Fine-Grained Network Telemetry System

Abstract

Network telemetry systems provide insights into network traffic that are essential for maintaining performance and security. While significant progress has been made in the area of telemetry system design and implementation, existing solutions focus on specific design aspects rather than providing a comprehensive end-to-end approach and require extensive expertise in data plane technology and programming. In response to these issues, this work presents IRIS: an interactive, end-to-end, and programmable fine-grained telemetry system built on the BroadScan module found in ASICs from Broadcom, Inc., which are deployed in a wide range of present-day networks. IRIS provides a flexible, high-level interface that enables users to define queries using a simple Python interface. It then translates the user-defined queries into hardware-level configurations for BroadScan, and efficiently retrieves the generated results from it. This work provides an overview of IRIS’s architecture and evaluates its ability to utilize BroadScan’s flow table at maximum capacity, produce time-based and loss-based statistics, and implement example use cases based on real-world network telemetry tasks.