This repository releases a dataset about the throughput and delivery time of adaptive video streaming, which was collected in real-world mobile networks from December 2019 to May 2021 and used in the following papers:
- Gerui Lv, Qinghua Wu, Weiran Wang, Zhenyu Li, and Gaogang Xie, "Lumos: towards Better Video Streaming QoE through Accurate Throughput Prediction," IEEE INFOCOM 2022.
- Gerui Lv, Qinghua Wu, Qingyue Tan, Weiran Wang, Zhenyu Li, and Gaogang Xie, "Accurate Throughput Prediction for Improving QoE in Mobile Adaptive Streaming," IEEE Transactions on Mobile Computing (TMC) 2023.
[2023-12-27 Update] Lumos training code is now available as ./lumos_train_dt.py. Due to the need for tuning and testing, there may be discrepancies between this version of the code and the descriptions in the paper. If so, please refer to the paper.
The dataset contains 590 video sessions running ABR algorithms (ABR_data, 392 sessions) and with constant bitrate level (constant_data, 198 sessions). Each session typically has a 5-minute or 5.5-minute playback duration. The following table provides detailed statistics. For more information, please refer to our paper.
| Type | Downstream Bandwidth | Connection Type | Signal Strength |
|---|---|---|---|
| ABR data (392) | 50Mbps: 198 5Mbps: 194 |
Wi-Fi: 237 4G: 155 |
Strong: 141 Medium: 114 Weak: 137 |
| Constant data (198) | 50Mbps: 120 5Mbps: 78 |
Wi-Fi: 144 4G: 54 |
Strong: 162 Medium: 6 Weak: 30 |
[Test time]-[Downstream bandwidth]-[Connection type]-[Signal strength]-[Video]-[ABR algorithm].csv
Test time: start time of a test (GMT+8); the format in Python strftime() and strptime() is "%y%m%d_%H%M"Downstream bandwidth: ["5Mbps", "50Mbps"], the limited bandwidth of the serverConnection type: ["wifi_2.4GHz", "wifi_5GHz", "4g"]Signal strength: ["strong", "medium", "weak"]Video: the format is [Video name]_[Chunk duration]; Video name is in ["bbb", "ed"] (Big Buck Bunny and Elephant Dream), and Chunk duration is in ["2s", "4s"]ABR algorithm: including RB, BBA (SIGCOMM '14), MPC and RobustMPC (SIGCOMM '15), Pensieve (SIGCOMM '17, retrained with our dataset), and HYB (described in Oboe, SIGCOMM '18)
Examples:
- 210415_1317-50Mbps-wifi_2.4GHz-medium-ed_4s-Pensieve.csv
- 210422_0115-5Mbps-wifi_5GHz-strong-bbb_4s-MPC.csv
- 210426_1225-50Mbps-4g-weak-bbb_4s-BBA.csv
Each .csv file corresponds to the data of a video session. The bitrate of each chunk is determined by the ABR algorithm. For each row from the 2nd line (the 1st line is the title filed) in the file, the format is as follows:
[downstream_bandwidth],[connection_type],[signal_strength],[bitrate(Kbps)],[chunk_size(KBytes)],[app_throughput(Kbps)],[delivery_time(s)],[player_state],[relative_chunk_index]
Examples:
- 5M,wifi,weak,4300,20137,6965,2.891,steady,1
- 50M,4g,strong,300,1223,7598,0.161,buffering,5
[Test time]-[Downstream bandwidth]-[Connection type]-[Signal strength]-[Video]-[Bitrate].csv
Test time,Downstream bandwidth,Signal strength, andVideoare the same as in ABR dataConnection type: ["wifi", "4g"]Bitrate: ["300Kbps", "750Kbps", "1200Kbps", "1850Kbps", "2850Kbps", "4300Kbps"]
Examples:
- 200426_1330-50Mbps-wifi-strong-bbb_4s-300Kbps.csv
- 200502_1107-50Mbps-4g-medium-bbb_4s-4300Kbps.csv
- 200504_1000-5Mbps-wifi-weak-bbb_4s-1200Kbps.csv
Each .csv file corresponds to the data of a video session at a constant bitrate. The data format is the same as in ABR data except for relative_chunk_index. In constant data, relative_chunk_index stays 0 for the steady state while increasing from 1 for the buffering state.
Examples:
- 50M,4g,weak,2850,15,116,0.125,steady,0
- 50M,wifi,strong,1200,6377,15746,0.405,buffering,1