Enhanced GNN generalization to Real Network Dataset by Attention Mechanism

1st place solution of Graph Neural Networking Challenge 2023

The problem

This challenge aims to create a Network Digital Twin based on neural networks that can accurately estimate QoS performance metrics given the network state and the input traffic. More in detail, solutions must predict the resulting per-flow mean delay given: (i) a network topology (L2 and L3), (ii) a set of input flow packet traces, and (iii) a routing configuration. The following figure presents a schematic representation.

Presentation and Awards Ceremony

Instructions to Replicate our Solution

To use this code, you should follow the tutorial described below.

Creating the Anaconda Virtual Environment

Considering that one has the Miniconda or the Anaconda virtual environment already installed in a Linux machine, to recreate the environment used in our solution, the below command should be used to create the environment:

conda env create -f environment.yml

After that, you can activate this environment using the next command:

conda activate gnn_2023_m0b1us

Training the m0b1us model

Considering the current directory is the m0b1us model, it is necessary to execute the command below to perform the training. The below command will train the CBR+MB model considering the dataset at datasets/data_cbr_mb_13. Where the --ckpt-path flag defines the weights directory name.

python3 std_train.py -ds CBR+MB --ckpt-path best_model

Performing the predictions

intended to create the predictions utilized in our best solution, the command below should be used. This command considers the CBR+MB model, the best weight 150-15.7196, the training and testing datasets at datasets/data_cbr_mb_13 and dataset/data_test, respectively.

python3 std_predict.py -ds CBR+MB --ckpt-path weights/150-15.7196 \
--tr-path datasets/data_cbr_mb_13_cv/0/training/ \
--te-path datasets/data_test/

Credits

This project would not have been possible without the contribution of:

Cláudio Matheus Modesto - LASSE, Federal University of Pará

Rebecca Aben-Athar - LASSE, Federal University of Pará

Andrey Silva - Ericsson

Silvia Lins - Ericsson