Chorus

This repository contains the experimental data and network traces for the paper:

Chorus: Coordinating Mobile Multipath Scheduling and Adaptive Video Streaming

Gerui Lv, Qinghua Wu, Yanmei Liu, Zhenyu Li, Qingyue Tan, Furong Yang, Wentao Chen, Yunfei Ma, Hongyu Guo, Ying Chen, Gaogang Xie

ACM MobiCom '24, November 18-22, 2024, Washington, D.C., USA

DOI: 10.1145/3636534.3649359

Overview

Chorus is a cross-layer framework that coordinates multipath scheduling with adaptive video streaming to jointly optimize Quality of Experience (QoE). Unlike traditional approaches that tune the packet scheduler independently, Chorus establishes two-way feedback control loops between the server transport layer and the client application, introducing Coarse-grained Decisions (CD) and Fine-grained Corrections (FC) to ensure appropriate bitrate selection and expected-time-oriented transport performance.

[Paper] [Slides] [Tech Report] [Demo Video] [GetMobile Highlights]

Code Availability

We released the Chorus Android Player, including the player-side DASH logic, the public transport service-provider interface, and a newly written Teki/JNI adapter skeleton with a fail-closed native stub. These sources expose the client/server QoE interface described in the technical report, but do not constitute a complete Chorus deployment.

Due to code ownership restrictions and legal reasons, the modified XQUIC transport core and headers, production backend/SDK artifacts, and server-side integration remain unavailable. This repository provides the main experimental datasets to facilitate reproducibility and further research.

Repository Structure

Chorus/
├── bw_traces/                  # Network bandwidth traces for emulation
├── emulation_results/          # Trace-driven emulation results (§5.2)
│   ├── Chorus/                 # Chorus results
│   ├── XLINK/                  # XLINK results
│   ├── MinRTT/                 # MinRTT results
│   ├── MinRTTRI/               # MinRTT+RI results
│   ├── SP/                     # Single-path (SP) results
│   └── *.csv                   # Summary metrics files
└── real-world_results/         # Real-world evaluation results (§5.5)
    ├── 1strong/                # Strong scenario
    ├── 2middle/                # Medium scenario
    └── 3weak/                  # Weak scenario

ABR Schemes

Scheme	Description
Chorus	Proposed cross-layer framework with CD&FC
XLINK	State-of-the-art QoE-driven MPQUIC scheduler (Baseline)
MinRTT	Basic MPQUIC scheduler, shortest RTT path selection (Baseline)
MinRTT+RI	MinRTT with unlimited Reinjection (upper bound of XLINK)
SP	Single-path QUIC (Baseline)

Dataset Details

Network Traces (bw_traces/)

Contains 52 network bandwidth traces collected from real 4G/5G cellular and WiFi networks:

File Type	Description
.pps	Packet-per-second traces for network emulation (Mahimahi/mpshell)
.csv	Time-series bandwidth data (Timestamp in seconds, Bandwidth in Mbps)
.png	Bandwidth trace visualizations
all_trace_bw.csv	Summary of all traces with average bandwidth, RTT, and loss rate

Trace naming: downlink-{scenario}.csv or uplink-{scenario}.csv

• Cellular scenarios: cellular_airport, cellular_driving, cellular_home, cellular_hsr (high-speed rail), cellular_office, cellular_outdoor, cellular_subway, etc.
• WiFi scenarios: wifi_home, wifi_office, wifi_walking, wifi_hsr

Emulation Results (emulation_results/)

Paper Section: Section 5.2 (Trace-driven Evaluation)

Each algorithm directory contains:

• {Algorithm}.test - Test configuration file
• summary_{Algorithm}.csv - Per-trace QoE results
• pred_summary_{Algorithm}.csv - Prediction metrics per trace
• tests/ - Raw per-test log files (converted to .csv)

Test group naming format (in tests/ directory):

{test_number}~{trace_name_path1}~{owd_path1}~{trace_name_path2}~{owd_path2}

Example: 1~cellular_subway_3~25~cellular_airport~35 represents:

• Test #1
• Path 1: cellular_subway_3 trace with 25ms OWD (one-way delay)
• Path 2: cellular_airport trace with 35ms OWD

.test file format:

Line 1: {algorithm} {duration_seconds}
Line 2: {path_type} {cc_count} {congestion_control}
        # MP = multipath, SP = single-path
Lines 3+: {trace1} {owd1} {loss1} {size1} {trace2} {owd2} {loss2} {size2}

Summary metrics files:

• overall_emulation_qoe_metrics.csv - QoE comparison (mean ± confidence interval)
• overall_emulation_mp_metrics.csv - Multipath transport metrics
• overall_emulation_predict_metrics.csv - Prediction accuracy metrics

Real-world Results (real-world_results/)

Paper Section: Section 5.5 (Real-world Evaluation)

108 field tests conducted on Android devices (4G and 5G smartphones) over commercial networks:

Scenario	Directory	Description
Strong	1strong/	WiFi bandwidth > 16 Mbps (highest bitrate)
Medium	2middle/	WiFi < 16 Mbps, but aggregated multipath sufficient
Weak	3weak/	Aggregated bandwidth often < 16 Mbps (critical scenario)

Per-test data structure: {scenario}/test{N}/{Algorithm}/

• {Algorithm}_client_{N}.csv - Client-side per-chunk metrics
• {Algorithm}_server_{N}.csv - Server-side metrics

Client CSV columns include:

• chunk_index, quality, bitrate, buffer_level
• delivery_time, rebuffer_time, QoE
• Throughput predictions: pre_throughput, hm_throughput
• Path metrics: fast_path_bw, slow_path_bw, fast_path_rtt, slow_path_rtt

Main Experimental Configuration

Parameter	Value
Video codec	H.264/MPEG-4
Chunk duration	4 seconds
Bitrate ladder	{1, 2.5, 5, 8, 16} Mbps
Resolution levels	360p, 480p, 720p, 1080p, 1440p (2K)
ABR algorithm	MPC (Model Predictive Control)
Congestion control	Cubic (decoupled per path)

Other settings are evaluated in §5.3 and §5.4. See the paper for details.

QoE Metric

Linear QoE formula (following MPC):

QoE = Σ Rₖ - μ·Σ Tₖ - λ·Σ |Rₖ₊₁ - Rₖ|

Where:

• Rₖ: Bitrate of chunk k (Mbps)
• Tₖ: Rebuffering time of chunk k (seconds)
• μ = 16: Rebuffering penalty coefficient (corresponding to the highest bitrate)
• λ = 1: Bitrate switch penalty coefficient

Citation

If you use this dataset in your research, please cite:

@inproceedings{lv2024chorus,
  title={Chorus: Coordinating Mobile Multipath Scheduling and Adaptive Video Streaming},
  author={Lv, Gerui and Wu, Qinghua and Liu, Yanmei and Li, Zhenyu and Tan, Qingyue and Yang, Furong and Chen, Wentao and Ma, Yunfei and Guo, Hongyu and Chen, Ying and Xie, Gaogang},
  booktitle={Proceedings of the 30th Annual International Conference on Mobile Computing and Networking (MobiCom '24)},
  pages={246--262},
  year={2024},
  publisher={ACM},
  address={New York, NY, USA},
  doi={10.1145/3636534.3649359}
}

Related Resources

• XQUIC: https://github.com/alibaba/xquic - User-space QUIC library used for Chorus implementation
• Mahimahi: https://github.com/ravinet/mahimahi - Network emulation tool for trace replay
• mpshell: https://github.com/ravinet/mahimahi/tree/old/mpshell_scripted - Multipath version of Mahimahi

License

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).