Network Digital Twin-Generated Dataset for Machine Learning-based Smart QoS-aware Zero-Touch Traffic Engineering

Overview

This record contains a set of synthetic datasets generated for the design and validation of intelligent, QoS-aware, zero-touch Traffic Engineering (TE) mechanisms. In this case, the dataset collects key metrics from each and every network link, including latency, jitter, packet loss, and input/output flows, within a realistic virtualized network environment.

To obtain this information without impacting the performance of the scenario, specific probes were deployed within the Network Digital Twin (NDT), capable of collecting metrics in a non-intrusive manner. In addition, data flows emulating congestion or degradation conditions were generated on the scenario, enriching the dataset and providing greater realism to the modeled situations.

To address this challenge, a Network Digital Twin (NDT) approach was employed to emulate realistic network conditions and traffic patterns, enabling the automated generation of labeled data that support the development of advanced adaptive and QoS-aware traffic engineering techniques.

Feature Set:

📌 General Information

Timestamp: time marker

Link: monitored network link

📌 Network Performance Metrics

Packet loss: percentage of lost packets

Jitter: variation in packet delay

RTT (Round-Trip Time): measured round-trip latency

TTL (Time to Live): hop-related metric

📌 Flow Metrics

Outflow_Router: outgoing traffic from the router

Inflow_Router: incoming traffic to the router

📌 Service Metrics

Resolve: DNS resolution success/failure indicator

Availability: service availability status

📌 Probe Metrics

Probe_duration: duration of the active probe/measurement

📌 Dataset Label

LABEL: 0/1/2

Dataset Variations:

To accommodate diverse research needs and scenarios, the dataset is provided in the following variations. Traffic bandwidth ranges emulated are as follows:

Range 0: 1–10 Mbps

Range 1: 11–40 Mbps

Range 2: 41–70 Mbps

Range 3: 71–90 Mbps

Range 4: Greater than 91 Mbps

dataset_01_TC33_27062025_labeled_wighted.csv

The bandwidth range distribution is approximately homogeneous, with around 20% for each generated range (ranges 0, 1, 2, 3, and 4), without applying packet loss or latency through "Traffic Control (TC)".

dataset_02_TC33_02072025_labeled_wighted.csv

The bandwidth range distribution is approximately homogeneous, with around 20% for each generated range (ranges 0, 1, 2, 3, and 4), applying packet loss between 0.1% and 1% on the output interfaces of the routers, as well as a delay of 10 ms ± 2 ms on the latency of that link.

dataset_03_TC33_05072025_labeled_wighted.csv

The bandwidth range distribution is approximately homogeneous, with around 20% for each generated range (ranges 0, 1, 2, 3, and 4), with TCP and UDP traffic applying packet loss between 0.1% and 1% on the output interfaces of the ingress, egress, and intermediate routers, as well as a delay of 10 ms ± 2 ms on the latency of that link.

dataset_01_TC33_27062025_labeled.csv

The bandwidth range distribution is approximately homogeneous, with around 20% for each generated range (ranges 0, 1, 2, 3, and 4), without applying packet loss or latency through "Traffic Control (TC)".

dataset_02_TC33_02072025_labeled.csv

The bandwidth range distribution is approximately homogeneous, with around 20% for each generated range (ranges 0, 1, 2, 3, and 4), applying packet loss between 0.1% and 1% on the output interfaces of the routers, as well as a delay of 10 ms ± 2 ms on the latency of that link.

dataset_03_TC33_05072025_labeled.csv

The bandwidth range distribution is approximately homogeneous, with around 20% for each generated range (ranges 0, 1, 2, 3, and 4), with TCP and UDP traffic applying packet loss between 0.1% and 1% on the output interfaces of the ingress, egress, and intermediate routers, as well as a delay of 10 ms ± 2 ms on the latency of that link.