Hello, if you have any need, please feel free to consult us, this is my wechat: wx91due

Project:

Reproduce a Network Research Result using Large Language Models

Problem Definition

In class, you have learned about a number of networking-related papers, but in the absence of open-source prototypes, manually reproducing one of these articles takes a long time. In this project, you will learn how to use large language models, chain-of-thought-based prompt engineering, and few-shot learning to reproduce networking papers. Your goals are to (1) select a networking domain paper and obtain a good grasp of the effects of large language models, and (2) reproduce your selected networking domain paper with our semi-automated reproduction framework and perform a evaluation.

Requirements

l Large Language Model: You can choose one or more large language models to help you reproduce the networking domain paper.

GPT-4.0

OpenAI o1

l Optional Papers：Choose one of the papers below to reproduce.

DCTCP: Data Center TCP

ABM: Active Buffer Management in Datacenters

GRoot: Proactive Verification of DNS Configurations

Tiramisu: Fast Multilayer Network Verification

Hedera: Dynamic Flow Scheduling for Data Center Networks

Arya: Arbitrary Graph Pattern Mining with Decomposition-based Sampling

l Metric: Criteria for judging the success of reproducing the paper.

Functional Evaluation: Verify if the replication system achieves the basic functionalities of the original system. If the replication system performs similarly to or better than the original system in these aspects, it can be deemed successful functionally.

Performance Evaluation: Compare the performance of the replication system with that of the original system. Performance metrics may include speed, stability, resource utilization, etc.

l Experimental Document: The information that needs to be recorded during the experiment.

Your choice of papers and large language model.

Number of Prompts : Count the number of all prompts you used, the number of prompts constructed with the semi-automated framework, the number of prompts used for debugging, and the number of prompts used in addition to semi-automated frameworks (human involvement).

Reproduction Time: Count the total time spent from the time you read the paper to the time you reproduced the system, and how much of that time was spent reading the paper, how much was spent on code generation, and how much was spent on debugging.

Realized Functions: What functions were implemented in the original system and what functions did you reproduce. For example:

If there are functions that have not been reproduced, please explain why.

Performance Comparison (One type of performance in one table): Perform two or more performance evaluations, with each performance evaluation tested with at least two datasets. As much as possible, use the dataset used by the original system. For example:

Time:

LOC (Line of Code): Count the lines of code of the reproduction system and original system.

Submission

Please submit by email to the instructor. Turn in electronic and paper material as follows.

Submission should include:

l Your source code.

l Links to conversations with LLM.

l A design document including evaluation.