A third year individual project.

Install Python 3.6.13

Install CUDA toolkit 10.2 (https://developer.nvidia.com/cuda-10.2-download-archive?target_os=Windows&target_arch=x86_64&target_version=10&target_type=exenetwork), and cudnn 8.1.1 (https://developer.nvidia.com/rdp/cudnn-download)

For cudnn installation guide, see: https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-windows

conda create --name edge-cloud-resource-allocation python=3.6

conda activate edge-cloud-resource-allocation

pip install --upgrade pip

pip install -r requirements.txt

For Windows

pip install -U https://s3-us-west-2.amazonaws.com/ray-wheels/latest/ray-2.0.0.dev0-cp36-cp36m-win_amd64.whl

For MacOS

pip install -U https://s3-us-west-2.amazonaws.com/ray-wheels/latest/ray-2.0.0.dev0-cp36-cp36m-macosx_10_15_intel.whl

• with GPU:

pip install torch==1.8.1+cu102 torchvision==0.9.1+cu102 torchaudio===0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

• without GPU:

pip install torch==1.8.1 torchvision==0.9.1 torchaudio===0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

• To install the CPLEX-Python modules on your system, use the script setup.py located in CPLEX_Studio1210/cplex/python/VERSION/PLATFORM.

module load python

python3 -m venv venv

source venv/bin/activate

pip install --upgrade pip

pip install -r requirements.txt

pip install -U https://s3-us-west-2.amazonaws.com/ray-wheels/latest/ray-2.0.0.dev0-cp36-cp36m-manylinux2014_x86_64.whl

sbatch --array=<config_indices> -p lyceum run_<experiment_name>.sh

(Optional) Use tmux: module load tmux then tmux

tensorboard --logdir="results/<experiment>"

or

tensorboard dev upload --logdir "results/<experiment>" --name "MARL Disaster Response" --description "Training drones in a gridworld disaster simulation environment"

(ctrl-b d to exit tmux window)

cd MARL-ReverseAuction/marl-edge-cloud
module load conda/py3-latest
source activate jack
bash ./src/train_MARL_mac.sh

cd MARL-ReverseAuction/marl-edge-cloud
module load conda/py3-latest
source activate jack
sinteractive -p gpu --gres=gpu:2 --time=04:00:00 --mem=16GB --job-name=marl-edge-cloud

then use it like a local machine, e.g.,

bash ./train_MARL_mac.sh

sbatch --array=<config_indices> -p batch run_<experiment_name>.sh or sbatch --array=<config_indices> -p gpu run_<experiment_name>.sh

First, add the following lines to your .ssh/config on your laptop. Look here if you want a slightly more detailed instruction on how to access your computer remotely.

Host workstation
     HostName <hostname>
     User <username>
     Port <port>
     LocalForward 8888 localhost:8888
     LocalForward 6006 localhost:6006

ssh -L 8000:localhost:8888 fb1n15@iridis5_b.soton.ac.uk

• You can use ssh -L to forward a port from your local machine to a remote machine.
• to run tensorboard, typing tensorboard --logdir="results/<experiment>" --port=6006 on the logging node (not on the iteractive job node). (6006 can be changed to any port number)
• source1, source2

• Do not load conda on the login node before requesting an interactive node. (This may cause the ModuleNotFoundError: No module named 'ray' error in the interactive node.)

• move iridis5 output file (slurm-***.out) to a folder (maybe called iridis5-output)
• how to put other parameters to the name of the RL results? So that I can distinguish different simulations.