Abstract

Traffic measurement is fundamental to network management, supporting key functions such as detecting large flows for improved load balancing, estimating flow sizes for better bandwidth allocation, and identifying suspicious traffic for intrusion prevention. As network traffic continues to grow, measurement systems must provide accurate flow statistics, such as frequency and persistence, at high speed and low latency, while operating efficiently on hardware platforms like programmable switches.

While deterministic approaches offer precision, they are often too slow and memory-intensive for real-time, large-scale deployments. Consequently, approximate methods have emerged as a practical alternative, striking a balance between accuracy and efficiency. In this talk, Weihe presents a series of novel approximate data structures (sketches) that enable fast and accurate flow detection across a variety of tasks in high-speed networks.