In this page, you can download the datasets which are used in our paper: Mariem BEN YAHIA, Yannick LE LOUEDEC, Gwendal SIMON, Loutfi NUAYMI, Xavier CORBILLON. HTTP/2-Based Frame Discarding for Low-Latency Adaptive Video Streaming. Submitted to ACM TOMM 2018

If you want to download the files, please provide a reference.

We provide mobile and WiFi network traces. To collect the real network traces of TCP traffic required for our simulations, we set up a platform composed of a CDN edge server from a French incumbent telecom operator and a WGET client running on an Ubuntu 16.04 machine. We achieve two series of 65-second long network captures and then we extract network captures of 6-second long. In the first series the client is connected to a 4G cellular network, and in the second series to the WiFi interface of an xDSL residential gateway. The captures are achieved with the Wireshark tool while the client is being downloading a file from the CDN server; the size of the file is big enough so that the download starts before and ends after the network capture.


The logs are formatted as follows:

TCP Packet Id

TCP packet sequence number

TCP packet length (Byte)

TCP packet reception time (second)

An example looks like this: 55.0 103897 1470.0 0.235646

Database: Mobile.txt (232 KB) This dataset presents mobile network traces

Database: WiFi.txt (318 KB) This dataset presents WiFi network traces

Throughput description

The throughput varies from 0.5 Mbps to 7 Mbps in the mobile network captures, and from 3 Mbps to 6 Mbps in the WiFi network captures. From each network capture we extract 6 different network traces of 6 seconds - which is the HAS segment duration of the videos considered in our simulation -, and with an average throughput of, respectively, 5 Mbps on the WiFi network, and 3.3 Mbps on the mobile network.

Mobile capture
WiFi capture