This project is aimed to studying the possibility of utilizing reinforcement learning to handle network congestion.
We use a version of TCP Cubic in which has it parameters exposed Beta, TCP Friendliness, Fast Convergence with an additional alpha parameter.
We utilize Proximal Policy Optimization to fine tune these parameters to handle congestion conditions.
- Ubuntu Server 16.04, Linux Kernel 4.4 (One for server and one for client)
- Machine must have at least 15 GB of storage and 2 GB of RAM
- OpenMPI
- Miniconda 3
- TCPTuner
- WonderShaper
- Git
- iperf3
- Update Ubuntu server
$ sudo apt update && sudo apt upgrade- Install Git, Linux kernel headers, Build-essentials, libsm6, libxrender1, libfontconfig1, OpenMPI & iperf3
$ sudo apt install git linux-headers-$(uname -r) build-essential libsm6 libxrender1 libfontconfig1 openmpi-bin openmpi-doc libopenmpi-dev iperf3- Install Miniconda and add Miniconda bin to path
$ cd ~
$ mkdir miniInstall
$ cd miniInstall
$ curl https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh --output Miniconda3-latest-Linux-x86_64.sh
$ chmod 755 Miniconda3-latest-Linux-x86_64.sh
$ ./Miniconda3-latest-Linux-x86_64.sh
$ echo 'PATH=$HOME/miniconda3/bin:$PATH' >> ~/.bashrc- Install TCPTuner and set congestion algorithm to tuner
$ cd ~
$ git clone https://github.com/Gasparila/TCPTuner
$ cd TCPTuner
$ cd module/
$ make
$ sudo rmmod tcp_tuner.ko
$ sudo insmod tcp_tuner.ko
$ sudo sysctl -w net.ipv4.tcp_congestion_control=tuner- Cloning repo
$ git clone ssh://git@github.com:Omar-Handouk/CMuRL_Implementation.git- Creating a new Anaconda Environment
$ conda create --name CMuRL python=3.7- Installing requirements
$ pip install -r requirments.txt- Run an iperf server on a remote machine
$ iperf3 -s- Run client on host machine
$ ./scripts/client.sh- Train the model and run it
$ python main.py