PINC (P4 and Intent for Network Configuration): A Prototyoe Using Large Language Models

Abstract

Networks require constant reconfiguration to meet changing demands, including adding security rules, optimizing traffic flow, and implementing new services. Currently, network operators must manually write complex code in specialized languages like P4, a time-consuming process requiring technical expertise. 

PINC (P4 and Intent for Network Configuration) explores whether artificial intelligence can bridge this gap. Instead of writing P4 code directly, operators describe what they want in plain language (for example, "block suspicious traffic from this region"), and the system automatically generates the necessary network programming. 

This research involves fine-tuning a large language model specialized for network programming and building a verification system that checks whether AI-generated P4 code is compliable. Through testing on logic and functional correctness, the system successfully generates compliable code up to 80-98% of the time, compared to just 10-36% with an unmodified AI model. 

The symposium poster will explain how the system is trained to understand operator intentions, the current verification process, and three key directions for future development.