Supporting Large Random Forests in the Pipelines of a Hardware Switch to Classify Packets at 100-Gbps Line Rate

Abstract

Packet classification is an essential function for many applications such as QoS provisioning and network intrusion detection. In this work, we perform random forest classification in the pipelines of a hardware switch to classify packets at 100 Gbps line rate. We design, implement, and evaluate the performance of our scheme in a P4 (Programming Protocol-independent Packet Processors) programmable hardware switch. Experimental results show that our scheme can: 1) support a random forest composed of more than 100 decision trees in the pipelines of a hardware switch; and 2) use such a large random forest to classify packets at 100 Gbps line rate. In this paper, we design the match-action rules that are required to implement such a random forest in a P4 hardware switch. Besides, we analytically derive the formulas that give the number of these rules.

Keywords

• Machine learning
• random forest
• packet classification
• programmable switches
• P4