0. Introduction

This repository contains the source code for our SIGCOMM'21 paper "Network Planning with Deep Reinforcement Learning".

Notes

The network topologies and the trained models used in the paper are not open-sourced. One can create synthetic topologies according to the problem formulation in the paper or modify the code for their own use case.

1. Environment config

AWS instance configurations

• AMI image: "Deep Learning AMI (Ubuntu 16.04) Version 43.0 - ami-0774e48892bd5f116"
• for First-stage: g4dn.4xlarge; Threads 16 in gurobi.env
• for others (ILP, ILP-heur, Second-stage): m5zn.12xlarge; Threads 8 in gurobi.env

Step 0: download the git repo

Step 1: install Linux dependencies

sudo apt-get update
sudo apt-get install build-essential libopenmpi-dev libboost-all-dev

Step 2: install Gurobi

cd <repo>/
./gurobi.sh
source ~/.bashrc

• Install the license here: https://www.gurobi.com/downloads/free-academic-license/
• Make sure your Gurobi solver work: gurobi_cl /opt/gurobi902/linux64/examples/data/coins.lp

Step 3: setup && start conda environment with python3.7.7

If you use the AWS Deep Learning AMI, conda is preinstalled.

conda create --name <env> python=3.7.7
conda activate <env>

Step 4: install python dependencies in the conda env

cd <repo>/spinninup
pip install -e .
pip install networkx pulp pybind11 xlrd==1.2.0

Step 5: compile C++ program with pybind11

cd <repo>/source/c_solver
./compile.sh

2. Content

• source
  • c_solver: C++ implementation with Gurobi APIs for ILP solver and network plan evaluator
  • planning: ILP and ILP-heur implementation
  • results: store the provided trained models and solutions, and the training log
  • rl: the implementations of Critic-Actor, RL environment and RL solver
  • simulate: python classes of flow, spof, and traffic matrix
  • topology: python classes of network topology (both optical layer and IP layer)
  • test.py: the main script used to reproduce results
• spinningup
  • adapted from OpenAI Spinning Up
• gurobi.sh
  • used to install Gurobi solver

3. Reproduce results (for SIGCOMM'21 artifact evaluation)

Notes

• Some data points are time-consuming to get (i.e., First-stage for A-0, A-0.25, A-0.5, A-0.75 in Figure 8 and B, C, D, E in Figure 9). We provide pretrained models in <repo>/source/results/trained/<topo_name>/, which will be loaded by default.
• We recommend distributing different data points and differetnt experiments on multiple AWS instances to run simultaneously.
• The default epoch_num for Figure 10, 11 and 12 is set to be 1024, to guarantee the convergence. The training process can be terminated manually if convergence is observed.

How to reproduce

• cd <repo>/source
• Figure 7: python test.py fig_7 <epoch_num>, epoch_num can be set smaller than 10 (e.g. 2) to get results faster.
• Figure 8: python test.py single_dp_fig8 <alg> <adjust_factor> produces one data point at a time (the default adjust_factor is 1).
  • For example, python test.py single_dp_fig8 ILP 0.0 runs ILP algorithm for A-0.
  • Pretrained models will be loaded by default if provided in source/results/trained/. To train from scratch which is NOT RECOMMENDED, run python test.py single_dp_fig8 <alg> <adjust_factor> False
• Figure 9&13: python test.py single_dp_fig9 <topo_name> <alg> produces one data point at a time.
  • For example, python test.py single_dp_fig9 E NeuroPlan runs NeuroPlan (First-stage) for topology E with the pretrained model. To train from scratch which is NOT RECOMMENDED, run python test.py single_dp_fig9 E NeuroPlan False.
  • python test.py second_stage <topo_name> <sol_path> <relax_factor> can load the solution from the first stage in <sol_path> and run second-stage with relax_factor=<relax_factor> on topo <topo_name>. For example, python test.py second_stage D "results/<log_dir>/opt_topo/***.txt" 1.5
  • we also provide our results of First-stage in results/trained/<topo_name>/<topo_name>.txt, which can be used to run second-stage directly. For example, python test.py second_stage C "results/trained/C/C.txt" 1.5
• Figure 10: python test.py fig_10 <adjust_factor> <num_gnn_layer>.
  • adjust_factor={0.0, 0.5, 1.0}, num_gnn_layer={0, 2, 4}
  • For example, python test.py fig_10 0.5 2 runs NeuroPlan with 2-layer GNNs for topology A-0.5
• Figure 11: python test.py fig_11 <adjust_factor> <mlp_hidden_size>.
  • adjust_factor={0.0, 0.5, 1.0}, mlp_hidden_size={64, 256, 512}
  • For example, python test.py fig_11 0.0 512 runs NeuroPlan with hidden_size=512 for topology A-0
• Figure 12: python test.py fig_12 <adjust_factor> <max_unit_per_step>.
  • adjust_factor={0.0, 0.5, 1.0}, max_unit_per_step={1, 4, 16}
  • For example, python test.py fig_11 1.0 4 runs NeuroPlan with max_unit_per_step=4 for topology A-1

4. Contact

For any question, please contact hzhu at jhu dot edu.