Leo: Online ML-based Traffic Classification at Multi-Terabit Line Rate

Explore a groundbreaking conference talk on Leo, a system for online traffic classification at multi-terabit line rates. Discover how this innovative approach implements an online machine learning model, specifically a decision tree, directly in the network switch's data plane. Learn about Leo's key features, including its ability to classify packets at switch line rate, automatically select resource-efficient designs, and update models on-the-fly without switch downtime. Delve into the implementation details on Intel Tofino switches and examine evaluation results showcasing Leo's performance in classifying traffic at line rate with minimal latency overhead. Gain insights into how Leo scales to larger model sizes compared to existing data plane ML classification systems while maintaining classification accuracy comparable to offline traffic classifiers. Understand the critical applications enabled by online traffic classification, such as network intrusion detection and prevention, Quality-of-Service provision, and real-time IoT analytics.