GenCC: Heterogeneous Network Congestion Control using LLMs
TL;DR
GenCC, a framework that leverages the code generation capabilities of large language models (LLMs), coupled with a realistic network testbed, to automatically synthesize congestion-control utility functions, demonstrates that LLMs provide a practical and effective approach to automating the design of high-performance congestion control protocols.
Abstract
Congestion control protocols regulate sending rates to optimize application performance and network utilization. In heterogeneous networks, however, applications often have different and conflicting performance objectives, making the design of suitable utility functions a challenging task that traditionally requires extensive mathematical analysis and experimental validation. We present GenCC, a framework that leverages the code generation capabilities of large language models (LLMs), coupled with a realistic network testbed, to automatically synthesize congestion-control utility functions. GenCC supports multiple guidance strategies, including mathematical chain-of-thought reasoning and evolutionary code refinement, enabling the generation of utility functions tailored to application requirements and network conditions. Our evaluation shows that LLM-generated utility functions can closely approach optimal performance and outperform the state-of-the-art heterogeneous congestion control protocol by 37% – 142%, depending on the scenario. These results demonstrate that LLMs provide a practical and effective approach to automating the design of high-performance congestion control protocols.