Projecte llegit
Títol: Scheduling in TSN networks using machine learning
Estudiants que han llegit aquest projecte:
MARTÍNEZ LOPERA, ARNAU (data lectura: 14-07-2023)- Cerca aquest projecte a Bibliotècnica
MARTÍNEZ LOPERA, ARNAU (data lectura: 14-07-2023)Director/a: AGUSTÍ TORRA, ANNA
Departament: ENTEL
Títol: Scheduling in TSN networks using machine learning
Data inici oferta: 23-09-2022 Data finalització oferta: 23-04-2023
Estudis d'assignació del projecte:
- GR ENG SIS TELECOMUN
- GR ENG TELEMÀTICA
| Tipus: Individual | |
| Lloc de realització: EETAC | |
| Paraules clau: | |
| TSN, scheduling, machine learning, KNN, SVM | |
| Descripció del contingut i pla d'activitats: | |
| Time Sensitive Networking (TSN) és el nom que reben
un conjunt de protocols que mantenen restriccions de temps real en xarxes de commutació de paquets (que, en general, no ho permeten per defecte). Un dels aspectes clau en xarxes TSN és utilitzar mecanismes d'assignació de recursos (scheduling) que proporcionin configuracions que permetin garantir els requisits temporals exigits pels fluxos que es volen transmetre. En aquest projecte es proposa programar, provar i avaluar mecanismes de Machine learning per tal de verificar si aquests mecanismes poden accelerar el procés d'obtenir configuracions vàlides que compleixen els requisits temporals necessaris. |
|
| Overview (resum en anglès): | |
| The massive adoption of Ethernet technology in multiple sectors, produces the need to provide deterministic solutions to ensure a Quality of Service (QoS) that meets the requirements of time-triggered flows. For this, the Time-Sensitive Networking (TSN) Task Group (TG) of the IEEE 802.1 developed a set of standards that define mechanisms for time-sensitive transmissions of data over Ethernet networks.
This project focuses on studying the feasibility of scheduling three classes of time-triggered flows with different time constraints over a simple network topology, which is made from two TSN (Time-Sensitive Networking) nodes connected through a link. Scheduling multiple time-triggered flows is a complex problem because the scheduling, if exists, must meet the time constraints of all these flows. To address this challenge, we explore the potential of using supervised machine learning classification models to accurately predict the feasibility of scheduling a given set of time-triggered flows, meeting their time-constraints, in a Time-Sensitive Network (TSN). Supervised models require a training dataset that contains a data matrix and a class label vector. To obtain the class label vector of each observation, we use an adaptation of the implementation developed in [27] of the Integer Linear Programming (ILP) model introduced in [33]. Two different models are considered: K-Nearest Neighbours (K-NN) and Support Vector Machine (SVM). These algorithms are tested and built from the application of the Leave One Out Cross-Validation (LOOCV) technique with the generated datasets, and the results obtained are compared and discussed. Finally, a hybrid verification strategy is proposed to train and test machine learning models, drastically reducing the resources and computation time originally required to compute the class label of each observation of the dataset. |
|