Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Toward Non-Expert Customized Congestion Control

Mingrui Zhang, Hamid Bagheri, Lisong Xu

TL;DR

NECC tackles the gap where general congestion control algorithms fail to meet individual user requirements by enabling non-experts to model, implement, and deploy customized CCAs using two LLMs and a BPF-based kernel attachment. The core approach is refinement-based code generation guided by domain-specific prompts and safety constraints, with iterative feedback to fix compilation, safety, and performance issues. Evaluation with real CCAs and live-streaming scenarios demonstrates improved streaming quality in congested networks and highlights the importance of CoT prompting and structured feedback. The work advances practical, user-centric network control by bridging high-level requirements and deployable kernel-space code, while outlining clear future research directions and limitations.

Abstract

General-purpose congestion control algorithms (CCAs) are designed to achieve general congestion control goals, but they may not meet the specific requirements of certain users. Customized CCAs can meet certain users' specific requirements; however, non-expert users often lack the expertise to implement them. In this paper, we present an exploratory non-expert customized CCA framework, named NECC, which enables non-expert users to easily model, implement, and deploy their customized CCAs by leveraging Large Language Models and the Berkeley Packet Filter (BPF) interface. To the best of our knowledge, we are the first to address the customized CCA implementation problem. Our evaluations using real-world CCAs show that the performance of NECC is very promising, and we discuss the insights that we find and possible future research directions.

Toward Non-Expert Customized Congestion Control

TL;DR

NECC tackles the gap where general congestion control algorithms fail to meet individual user requirements by enabling non-experts to model, implement, and deploy customized CCAs using two LLMs and a BPF-based kernel attachment. The core approach is refinement-based code generation guided by domain-specific prompts and safety constraints, with iterative feedback to fix compilation, safety, and performance issues. Evaluation with real CCAs and live-streaming scenarios demonstrates improved streaming quality in congested networks and highlights the importance of CoT prompting and structured feedback. The work advances practical, user-centric network control by bridging high-level requirements and deployable kernel-space code, while outlining clear future research directions and limitations.

Abstract

General-purpose congestion control algorithms (CCAs) are designed to achieve general congestion control goals, but they may not meet the specific requirements of certain users. Customized CCAs can meet certain users' specific requirements; however, non-expert users often lack the expertise to implement them. In this paper, we present an exploratory non-expert customized CCA framework, named NECC, which enables non-expert users to easily model, implement, and deploy their customized CCAs by leveraging Large Language Models and the Berkeley Packet Filter (BPF) interface. To the best of our knowledge, we are the first to address the customized CCA implementation problem. Our evaluations using real-world CCAs show that the performance of NECC is very promising, and we discuss the insights that we find and possible future research directions.
Paper Structure (30 sections, 6 figures, 2 tables)

This paper contains 30 sections, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem statement
  4. Proposed Framework
  5. Framework Overview
  6. Design Choices for Achieving Design Goals
  7. Refinement-Based CCA Generation
  8. Prompt of Implementation LLM
  9. Instructions
  10. CCA Design Information
  11. R1: Minimum throughput requirement
  12. R2: Maximum throughput requirement
  13. R3: Mandatory throughput reduction with persistent loss
  14. Additional Information
  15. Feedback to Implementation LLM
  16. ...and 15 more sections

Figures (6)

  • Figure 1: Framework of Non-Expert Customized CCA
  • Figure 2: Code quality CDF in different temperatures: temperature 0.5 performs better than temperature 1.
  • Figure 3: CoT may or may not improve code quality compared with 0-shot prompting, mainly due to compilation errors.
  • Figure 4: A larger pool size improves the highest satisfaction score of the pool, but it also leads to a higher cost.
  • Figure 5: Feedback effectively improves the code quality over iterations.
  • ...and 1 more figures