Dataset of a 5G RTSP video streaming use case

About the project: monitoring 5G RTSP video streaming

This dataset collects data from a 5G video streaming use case. A video is streamed by a cvlc server (realized as a Kubernetes pod) through RTSP to a variable number of 5G UE clients that activate according to a daily traffic pattern. The values of the 4 dataset features (number of active UEs, gNB's downlink bit rate, pod's outbound traffic, and pod's CPU usage) are collected by a custom monitoring system deployed in the context of the MONB5G project.

Setup/Equipment

The Kubernetes cluster, including the server pod, runs in a COTS server. The 5G core and gNB is realized through Amarisoft Callbox Ultimate. The UEs are emulated through Amarisoft Simbox. In order to display the video in ffplay clients, we use the Remote UE from Amarisoft, so traffic from Simbox is forwarded to an external VM with GUI.

Video

The streamed video is Big Buck Bunny at 30 FPS from https://peach.blender.org/.

Video codec 

Advanced Video Codec (AVC) 

Width 

1920 pixels 

Height 

1080 pixels 

Display aspect radio 

16:9 

Duration 

10 min 34 s 

Max Bitrate 

16.7 Mb/s 

Frame rate 

30 FPS 

What does this Zenodo project contain?

The csv file of the dataset (dataset.csv)

A picture displaying an overview of the setup (overview.png)

A picture displaying Grafana charts for each featuer (grafana.png)

A picture displaying a screenshot of the Remote UE VM with multiple UEs playing the video (ues.png)

Dataset

The dataset has 5 columns (time + 4 features). Features:

Time: timestamp in epoch format.

Number of active UEs (N): number of UEs that are currently downloading more than 100 kbps. No unit.

gNB's downlink bit rate (R): aggregate downlinkg bitrate from the gNB to all the UEs. In Mbps.

Outbound traffic (O): outboun traffic at the pod's interface, transmitting the video(s) packets. In Mbps.

CPU (C): CPU usage at the server pod. In millicores (mc). Each iteration represents a whole day, composed of 24 "demand periods". Each demand period takes 2 minutes and is given by the number of active UEs consuming the video stream (N). N is included for informative reasons. Sampling rate is 10 seconds, but some parameters are refreshed at a lower frequency given monitoring limitations. This means that some parameters repeat the same value in consecutive measurements.