ACM Multimedia 2019 Grand Challenge: Live Video Streaming

This repository contains the simulator code being used for ACM Multimedia 2019 Live Video Streaming. The simulator is a frame-level DASH video streaming simulator with a pluggable client module for bitrate control and latency control.

Files

The simulator contains the following files:

• run.py : An SDK to call the SIM and ABR
• ABR.py : your ABR algorithm
• fixed_env.py : SIM code simulates live streaming player download, play, card, skip frame, etc
• load_trace.py : load trace to memory

The folder dataset contains the network traces and video traces. Please see the README.md file in the dataset directory for decsription of the dataset.

The Simulator

The simulator simulates a live video player，including downloading and playback of video frames. It reads as inputs:

1. A video trace, which contains the size of each video frame in a video file.
2. A network trace, which contains the network throughput at each time instance.
3. The bitrate control and latency control algorithm provided by participant.

The simulator output the following:

params	params description	example
time(s)	physical time	0.46(s)
time_interval(s)	duration in this cycle	0.012(s)
send_data_size(bit)	The data size downloaded in this cycle	14871(b)
frame_time_len(s)	The time length of the frame currently	0.04(s)
rebuf(s)	The rebuf time of this cycle	0.00(s)
buffer_size(s)	The buffer size time length	1.26(s)
play_time_len(s)	The time length of playing in this cycle	0.012(s)
end_delay(s)	Current end-to-to delay	1.31(s)
cdn_newest_id	Cdn the newest frame id	85
download_id	Download frame id	41
cdn_has_frame	cdn cumulative frame info	1.31(s)
decision_flag	Gop boundary flag or I frame flag	False
buffer_flag	Whether the player is buffering	False
cdn_rebuf_flag	Whether the cdn is rebuf	False
end_of_video	Whether the end of video	False

Requirement

You will need Python 3 to run the simulator.

Running the Simulator

To run the simulator, you execute

python run.py

The given default code should produce something like the following:

network traceID, network_reward, avg_running_time 1 641.1484318760413 2.7569223579257976e-06
network traceID, network_reward, avg_running_time 2 985.7361064180305 1.569810351953533e-06
network traceID, network_reward, avg_running_time 3 1434.0326303686352 1.506220995549704e-06
network traceID, network_reward, avg_running_time 4 909.2318048157695 1.5385812629050623e-06
network traceID, network_reward, avg_running_time 5 635.4616496246817 2.8796182701514625e-06
network traceID, network_reward, avg_running_time 6 1019.0324002078656 3.573290152802109e-06
network traceID, network_reward, avg_running_time 7 991.7406496996467 4.670511970121271e-06
network traceID, network_reward, avg_running_time 8 1055.3980756167875 3.7547605615472396e-06
network traceID, network_reward, avg_running_time 9 -163.7251464104959 3.7670135498046877e-06
network traceID, network_reward, avg_running_time 10 460.11141008945896 1.8877298097065138e-06
network traceID, network_reward, avg_running_time 11 1586.2008041285537 1.5093755589222177e-06
network traceID, network_reward, avg_running_time 12 1028.6844963417639 1.4955950978738685e-06
network traceID, network_reward, avg_running_time 13 1021.9785822724941 1.5320970756073211e-06
network traceID, network_reward, avg_running_time 14 963.8776511160278 2.4441224950933854e-06
network traceID, network_reward, avg_running_time 15 81.1547975421813 3.175536099211264e-06
network traceID, network_reward, avg_running_time 16 -19.11784167374209 2.672614120855564e-06
network traceID, network_reward, avg_running_time 17 1202.1056307541305 2.4344927753246596e-06
network traceID, network_reward, avg_running_time 18 442.34703880371285 3.3536329242844434e-06
network traceID, network_reward, avg_running_time 19 1009.585542619389 2.43250041951044e-06
network traceID, network_reward, avg_running_time 20 -307.0479419414425 2.85270988858062e-06
[748.8968386134745, 2.5903567868865584e-06]

The output above shows the resulting QoE value and your ABR algorithm's average running time for each of the 20 runs, each using a different network trace. The last line shows the average QoE and average running time for all 20 runs.

Configuring the Simulator

Participants can configure the simulator by editing run.py and set the following parameters.

• To change the video trace used in the simulation, you can change the variable VIDEO_TRACE. The default VIDEO_TRACE is set to AsianCup_China_Uzbekistan. Other possibilities are: Fengtimo_2018_11_3, YYF_2018_08_12, game, room, and sports.

• To change the network trace used in the simulation, you can change the variable NETWORK_TRACE. The default NETWORK_TRACE can be set to fixed, high, medium, or low.

• The simulator can produce detailed log files for debugging. To turn this on, set the variable DEBUG to True. By default, the logs will be written to a sub-directory called log. This log directory can be changed by setting the LOG_FILE_PATH variable. Note that you may want to set DEBUG to False if you are training an AI model as large volume of data may be written to disk when logging is on.