The following Readme is taken from the original work that provided the implementation to run experiments. The Technical report pdf and the paper have a detailed explanation of the changes done to the original implementation. Also, the docker folder on this project is our adaptation to the original implementation and has all the changes. The way to run it is described in the Technical Report and can be supported by the original work documentation.
This project contains my master thesis and all the files for reproducing the results.
This work is split in several public repositories:
- Multi-path scheduling with deep reinforcement learning: this repository. Includes the thesis, the offline RL agent, pre-compiled binaries and the testing scripts. Also, a docker image is available for the tests.
- GoRL: Open implementation of an agent in Go, as is described in the thesis.
- MP-QUIC: Fork of the multi-path QUIC implementation that includes the RL agent
Follow this instructions for installing all the testing framework of this thesis.
Download the last ubuntu server LTS (18.04 at this moment) and create a new VM on virtual box with the image. The main characteristics that I used are: 2 vCPU, 2048 MB of RAM and 20 GB of storage.
Once Ubuntu is installed, install docker:
$ sudo snap install docker
The pdf file Appendix-Technical Report has a deep description an explanation not only in the changes to the original code to achieve our work, but also a description of the meaning and purpose of the other repositories that are used in this project and the original code.