Deep Learning to Predict the Feasibility of Priority-Based Ethernet Network Configurations

Abstract

Machine learning has been recently applied in real-time systems to predict whether Ethernet network configurations are feasible in terms of meeting deadline constraints without executing conventional schedulability analysis. However, the existing prediction techniques require domain expertise to choose the relevant input features and do not perform consistently when topologies or traffic patterns differ significantly from the ones in the training data. To overcome these problems, we propose a Graph Neural Network (GNN) prediction model that synthesizes relevant features directly from the raw data. This deep learning model possesses the ability to exploit relations among flows, links, and queues in switched Ethernet networks and generalizes to unseen topologies and traffic patterns. We also explore the use of ensembles of GNNs and show that it enhances the robustness of the predictions. An evaluation on heterogeneous testing sets comprising realistic automotive networks shows that ensembles of 32 GNN models feature a prediction accuracy ranging from 79.3% to 90% for Ethernet networks using priorities as the Quality-of-Service mechanism. The use of ensemble models provides a speedup factor ranging from 77 to 1,715 compared to schedulability analysis, which allows a far more extensive design space exploration.

Exploring Trade-offs in Concept Design of Integrated Modular Avionic Platform Configurations : Topology Generation, Resource Adequacy, and Dependability

“…Long and Navet [62], for example, present a graph neural network approach to analyze the feasibility of Ethernet network configurations in the context of timesensitive networks. They show that an ensemble of 32 models provides higher prediction accuracy and speedup than traditional schedulability analysis.…”

Enhanced TSN Configuration Detection Using Optimized BPNN with Feature Selection for Industry 4.0

“…𝐿 = 𝑖𝑛 * ℎ𝑖𝑑𝑒 + ℎ𝑖𝑑𝑒 + ℎ𝑖𝑑𝑒 * 𝑜𝑢𝑡 (12) The initiation stage of the particle swarm optimization (PSO) process is pivotal for setting the foundational parameters that govern the behavior of the algorithm. Commence by establishing a predetermined population size that reflects the number of particles within the swarm.…”

“…Literature [21] explored the use of standard supervised and unsupervised machine learning techniques, demonstrating that KNN and K-means [22] can serve as alternatives for schedulability analysis. Literature [12] introduced a graph neural network (GNN) model, which exhibits a certain degree of robustness to changes in the network's topology and traffic patterns. However, the classification accuracy of this model ranges from 79.3% to 95.4%, indicating variability in accuracy and hindering its practical applicability in some cases.…”

“…However, the classification accuracy of this approach requires improvement. Additionally, literature [12] explored the use of graph neural network (GNN) model [13] , which demonstrated a degree of robustness to changes in the topology and traffic patterns of the training set, while the generalized classification accuracy is relatively low.…”