QoE-driven Joint Decision-Making for Multipath Adaptive Video Streaming

This repository contains the code for the submission "QoE-driven Joint Decision-Making for Multipath Adaptive Video Streaming" to Globecom 2023.

Steps to reproduce the results

Compile and install

Prerequisites

Only tested on a clean installation of Ubuntu server 22.04.1 LTS.

Install Mininet, according to the native installation from source guide.

cd ~
git clone https://github.com/mininet/mininet.git
cd mininet && git checkout -b 2.3.1b4

./util/install.sh -s 2.3.1b4 -a

Our experiments are conducted with Mininet 2.3.1b4.

Install dependencies

sudo apt-get update
sudo apt-get install wget curl build-essential git cmake unzip tree screen pkg-config libssl-dev libbrotli-dev mininet python3-full python3-virtualenv python3-pip -y

Compile

mkdir -p ~/clarkzjw-globecom23 && cd ~/clarkzjw-globecom23

git clone https://github.com/h2o/picotls.git
git clone --branch globecom2023 https://github.com/clarkzjw/picoquic.git
git clone --branch globecom2023 https://github.com/clarkzjw/globecom2023.git

cd picotls && git submodule init && git submodule update && cmake -DCMAKE_C_FLAGS="-fPIC" . && make
cd ../picoquic && cmake -DCMAKE_C_FLAGS="-fPIC" . && make

cd ../globecom2023 && virtualenv .venv --python=python3 && source .venv/bin/activate
pip install -r requirements.txt

cmake . && make

Dataset

Use our prebuilt dataset

Download our dataset, which takes about 20GB.

cd ~/clarkzjw-globecom23/
wget -O mpd.zip "https://zenodo.org/records/8346996/files/mpd.zip?download=1"
unzip mpd.zip

The sha256sum of the dataset file mpd.zip is shown below.

95a58132043993a6382f3bdd10fd465cea4c6f565bc243669c00f26c5b6cc0e1  mpd.zip

Create custom DASH dataset

Follow the instructions in the dataset folder to create your own DASH dataset.

But then, you have to update bitrate_mapping in bitrate.py with the custom bitrate ladder accordingly.

Generate certificates for QUIC

cd ~/clarkzjw-globecom23/globecom2023

openssl req -nodes -x509 -newkey rsa:2048 -days 365 -keyout ca-key.pem -out ca-cert.pem

Run Mininet emulations

sudo bash run.sh

The experiments will run using screen with root user. You can use sudo screen -ls and sudo screen -r to attach to the corresponding running session.

Generate figures

Generate corresponding figures in the paper

Run gen_figure.m within ./figure/gcloud and ./figure/mininet using MATLAB to generate the figures in the paper.

Run experiments on real-world network

The Terraform scripts in ./terraform are used in our experiment to create a multipath testbed on GCP's us-west-1a zone which is geographically closest to our location.

The traceroute results below shows there are two distinct paths from our location to the VM, which can be simplified with

traceroute to 34.105.96.243 (34.105.96.243), 30 hops max, 60 byte packets
1  _gateway (192.168.0.254)  0.539 ms  0.462 ms  0.426 ms
2  142.104.68.1 (142.104.68.1)  0.913 ms  0.873 ms  0.839 ms
3  142.104.124.105 (142.104.124.105)  1.080 ms  1.047 ms  1.017 ms
4  142.104.100.241 (142.104.100.241)  0.977 ms  1.065 ms  1.023 ms
5  cle-core-edge.bb.uvic.ca (142.104.100.189)  1.308 ms  1.270 ms  1.236 ms
6  207.23.244.233 (207.23.244.233)  1.199 ms  1.286 ms  1.179 ms
7  vctr3rtr2.network.canarie.ca (199.212.24.98)  1.287 ms  1.405 ms  1.669 ms
8  sttl1rtr2.canarie.ca (206.81.80.189)  3.596 ms  3.493 ms  3.597 ms
9  google-2-lo-std-707.sttlwa.pacificwave.net (207.231.242.22)  10.022 ms  11.144 ms  10.892 ms
10  243.96.105.34.bc.googleusercontent.com (34.105.96.243)  9.987 ms  9.449 ms  8.410 ms

traceroute to 34.105.96.243 (34.105.96.243), 30 hops max, 60 byte packets
 1  DD-WRT (192.168.1.1)  0.250 ms  0.248 ms  0.309 ms
 2  100.64.0.1 (100.64.0.1)  47.095 ms  68.122 ms  68.095 ms
 3  172.16.251.66 (172.16.251.66)  68.104 ms  68.076 ms  68.048 ms
 4  * * *
 5  undefined.hostname.localhost (206.224.64.13)  68.006 ms undefined.hostname.localhost (206.224.64.37)  67.978 ms undefined.hostname.localhost (206.224.64.13)  67.946 ms
 6  142.250.170.144 (142.250.170.144)  67.933 ms 142.250.163.222 (142.250.163.222)  65.939 ms 142.250.170.144 (142.250.170.144)  65.821 ms
 7  243.96.105.34.bc.googleusercontent.com (34.105.96.243)  76.793 ms  40.952 ms  70.777 ms

But our code can be deployed on any suitable multipath testbeds with the following requirements.

The following variables in downloader.py have to be updated accordingly.

`if_name_mapping`
`default_mpd_url`
`default_host`
`default_port`

and run main.py with custom options.

$ python3 main.py --help   

usage: main.py [-h] [--exp_id EXP_ID] [--scheduler {roundrobin,minrtt,contextual_bandit}] [--algorithm {LinUCB,LinTS,LinGreedy}] [--linucb_alpha LINUCB_ALPHA] [--lints_alpha LINTS_ALPHA]
               [--egreedy_epsilon EGREEDY_EPSILON] [--nb_segment NB_SEGMENT]

Adaptive video streaming with contextual bandits and MPQUIC

options:
  -h, --help            show this help message and exit
  --exp_id EXP_ID       experiment id
  --scheduler {roundrobin,minrtt,contextual_bandit}
                        scheduler
  --algorithm {LinUCB,LinTS,LinGreedy}
                        scheduling algorithm
  --linucb_alpha LINUCB_ALPHA
                        alpha for LinUCB
  --lints_alpha LINTS_ALPHA
                        alpha for LinTS
  --egreedy_epsilon EGREEDY_EPSILON
                        epsilon for epsilon-greedy
  --nb_segment NB_SEGMENT
                        number of segments to download

The metric results will be saved in ./result/ in json format and can be used to generate figures using scripts in analysis.

clarkzjw / globecom2023