Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RouteNet-Fermi: Network Modeling With GNN (Analysis And Re-implementation)

Shourya Verma, Simran Kadadi, Swathi Jayaprakash, Arpan Kumar Mahapatra, Ishaan Jain

TL;DR

RouteNet-Fermi addresses the challenge of predicting network performance in complex, large-scale networks using a Graph Neural Network that models interactions among flows, queues, and links. This work extends RouteNet-Fermi with LSTM and simple RNN variants alongside the original GRU, enabling direct comparisons of recurrent architectures within the GNN framework. The contributions include an extended, flexible implementation, evaluation across Fat Tree, real traffic, and traffic-model datasets, and analysis of memory, gradient flow, and throughput trade-offs in recurrent cells. The results indicate that LSTM generally yields strongest accuracy for delay and traffic patterns, GRU offers a robust efficiency-accuracy balance, and RNNs struggle with large, bursty networks, informing design choices for data-driven network modeling.

Abstract

Network performance modeling presents important challenges in modern computer networks due to increasing complexity, scale, and diverse traffic patterns. While traditional approaches like queuing theory and packet-level simulation have served as foundational tools, they face limitations in modeling complex traffic behaviors and scaling to large networks. This project presents an extended implementation of RouteNet-Fermi, a Graph Neural Network (GNN) architecture designed for network performance prediction, with additional recurrent neural network variants. We improve the the original architecture by implementing Long Short-Term Memory (LSTM) cells and Recurrent Neural Network (RNN) cells alongside the existing Gated Recurrent Unit (GRU) cells implementation. This work contributes to the understanding of recurrent neural architectures in GNN-based network modeling and provides a flexible framework for future experimentation with different cell types.

RouteNet-Fermi: Network Modeling With GNN (Analysis And Re-implementation)

TL;DR

RouteNet-Fermi addresses the challenge of predicting network performance in complex, large-scale networks using a Graph Neural Network that models interactions among flows, queues, and links. This work extends RouteNet-Fermi with LSTM and simple RNN variants alongside the original GRU, enabling direct comparisons of recurrent architectures within the GNN framework. The contributions include an extended, flexible implementation, evaluation across Fat Tree, real traffic, and traffic-model datasets, and analysis of memory, gradient flow, and throughput trade-offs in recurrent cells. The results indicate that LSTM generally yields strongest accuracy for delay and traffic patterns, GRU offers a robust efficiency-accuracy balance, and RNNs struggle with large, bursty networks, informing design choices for data-driven network modeling.

Abstract

Network performance modeling presents important challenges in modern computer networks due to increasing complexity, scale, and diverse traffic patterns. While traditional approaches like queuing theory and packet-level simulation have served as foundational tools, they face limitations in modeling complex traffic behaviors and scaling to large networks. This project presents an extended implementation of RouteNet-Fermi, a Graph Neural Network (GNN) architecture designed for network performance prediction, with additional recurrent neural network variants. We improve the the original architecture by implementing Long Short-Term Memory (LSTM) cells and Recurrent Neural Network (RNN) cells alongside the existing Gated Recurrent Unit (GRU) cells implementation. This work contributes to the understanding of recurrent neural architectures in GNN-based network modeling and provides a flexible framework for future experimentation with different cell types.

Paper Structure

This paper contains 19 sections, 11 equations, 15 figures, 5 tables.

Table of Contents

  1. Abstract
  2. Introduction
  3. Background
  4. Traditional Approaches
  5. RouteNet-Fermi Framework
  6. Dataset
  7. Scheduling Policies Datasets
  8. Scalability Dataset
  9. Real Traffic and Traffic Models Datasets
  10. Network Tasks
  11. Deep Learning Methods
  12. Training Configuration
  13. Results
  14. Fat Tree
  15. Real Traffic and Scalability
  16. ...and 4 more sections

Figures (15)

  • Figure 1: Black-box representation of RouteNet-Fermi
  • Figure 2: Schematic representation of RouteNet-Fermi
  • Figure 3: Comparison of neural network cell architectures. (a) RNN cell showing simple structure with a single hidden state. (b) LSTM cell with input, forget, output gates and memory cell state. (c) GRU cell with reset and update gates.
  • Figure 4: Comparison of model validation loss per epoch for Fat Tree topology for delay prediction
  • Figure 5: Comparison of model validation loss per epoch for Real Traffic topology and Scalability
  • ...and 10 more figures