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

in ACM Transactions on Cyber-Physical Systems (2021), 5(4), 126

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 ... [more ▼]

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 features 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 1715 compared to schedulability analysis, which allows a far more extensive design space exploration. [less ▲]

A Hybrid Machine Learning and Schedulability Method for the Verification of TSN Networks

in 15th IEEE International Workshop on Factory Communication Systems (WFCS2019) (2019, March)

Machine learning (ML), and supervised learning in particular, can be used to learn what makes it hard for a network to be feasible and try to predict whether a network configuration will be feasible ... [more ▼]

Machine learning (ML), and supervised learning in particular, can be used to learn what makes it hard for a network to be feasible and try to predict whether a network configuration will be feasible without executing a conventional schedulability analysis. A disadvantage of ML-based timing verification with respect to schedulability analysis is the possibility of "false positives": configurations deemed feasible while they are not. In this work, in order to minimize the rate of false positives, we propose the use of a measure of the uncertainty of the prediction to drop it when the uncertainty is too high, and rely instead on schedulability analysis. In this hybrid verification strategy, the clear-cut decisions are taken by ML, while the more difficult ones are taken by a conventional schedulability analysis. Importantly, the trade-off achieved between prediction accuracy and computing time can be controlled. We apply this hybrid verification method to Ethernet TSN networks and obtain, for instance in the case of priority scheduling with 8 traffic classes, a 99% prediction accuracy with a speedup factor of 5.7 with respect to conventional schedulability analysis and a reduction of 46% of the false positives compared to ML alone. [less ▲]

On the use of supervised machine learning for assessing schedulability: application to Ethernet TSN

in 27th International Conference on Real-Time Networks and Systems (RTNS 2019) (2019)

In this work, we ask if Machine Learning (ML) can provide a viable alternative to conventional schedulability analysis to determine whether a real-time Ethernet network meets a set of timing constraints ... [more ▼]

In this work, we ask if Machine Learning (ML) can provide a viable alternative to conventional schedulability analysis to determine whether a real-time Ethernet network meets a set of timing constraints. Otherwise said, can an algorithm learn what makes it difficult for a system to be feasible and predict whether a configuration will be feasible without executing a schedulability analysis? To get insights into this question, we apply a standard supervised ML technique, k-nearest neighbors (k-NN), and compare its accuracy and running times against precise and approximate schedulability analyses developed in Network-Calculus. The experiments consider different TSN scheduling solutions based on priority levels combined for one of them with traffic shaping. The results obtained on an automotive network topology suggest that k-NN is efficient at predicting the feasibility of realistic TSN networks, with an accuracy ranging from 91.8% to 95.9% depending on the exact TSN scheduling mechanism and a speedup of 190 over schedulability analysis for 10^6 configurations. Unlike schedulability analysis, ML leads however to a certain rate "false positives'' (i.e., configurations deemed feasible while they are not). Nonetheless ML-based feasibility assessment techniques offer new trade-offs between accuracy and computation time that are especially interesting in contexts such as design-space exploration where false positives can be tolerated during the exploration process. [less ▲]