MITuna

Tuna is a distributed tuning infrastructure that provides pre-compiled kernels for MIOpen customers through automated Jenkins pipelines and SLURM scalable architecture. MITuna also provides a scalable task management infrastructure ready to integrate with external libaries.


License
MIT
Install
pip install MITuna==0.1

Documentation

TUNA

Tuna is a distributed tuning infrastructure that provides tuning through automated Jenkins pipelines and SLURM scalable architecture. For MIOpen customers we provide pre-compiled kernels.

Prerequisites

Install python3.9

apt-get update && apt-get install software-properties-common
add-apt-repository ppa:deadsnakes/ppa
apt install python3.9

Install pip for python3.9

wget https://bootstrap.pypa.io/get-pip.py -o get-pip.py
python3.9 get-pip.py
rm get-pip.py

Install MySQL server

apt-get install mysql-server

Enable the service

systemctl start mysql

Install ipmitool

apt-get install ipmitool

Setup passwordless ssh between all machines, for example:

ssh-keygen -t rsa
ssh-copy-id <user>@<ip-address>
ssh <user>@<ip-address>

For the tuning cycle, every machine needs to be able to access every other machine through passwordless ssh.

Installation

Clone the repo using

git clone <repo url>

Enter the Tuna directory

cd MITuna

Create a virtual envornment, and activate it (by sourcing its activate script)

virtualenv -p python3.9 myvenv
source myvenv/bin/activate

Install the required dependencies:

python3.9 -m pip install -r requirements.txt

The above assumes that Tuna lives in the home directory and the virtual environment was created using the command indicated above.

Add the following environment variables to a local file and then source the file:

export TUNA_DB_USER_NAME=root
export TUNA_DB_PASSWORD=<password for root>
export TUNA_DB_HOSTNAME=localhost
export TUNA_DB_NAME=<database_name>
export gateway_ip=<gateway_ip>
export gateway_port=<gateway_port>
export gateway_user=<gateway_user>

All machines used in the tuning process must have ssh-keys enabled. MITuna needs to have all-to-all machine communication available and passwords must not be required at run-time.

Run the setup scripts:

python3.9 setup.py develop

The root tuna folder needs to be appended to the PYTHONAPTH:

export PYTHONPATH=/<path_to_MITuna>/:$PYTHONPATH

To create the database run the following script:

./tuna/miopen/db_tables.py

The installation and setup are now complete. To start a tuning cycle, please follow the steps documented in TuningCycle

Code formatting

MITuna used yapf for code formatting:

yapf -i --style='{based_on_style: google, indent_width: 2}' --recursive tuna/
yapf -i --style='{based_on_style: google, indent_width: 2}' --recursive tests/

Static code analysis

In order for a PR to be accepted the following pylint command needs to result in 10/10 analysis:

cd MITuna/tuna
pylint -f parseable -d duplicate-code --max-args=8 --indent-string '  ' *.py

Coverage

To extract coverage percentage from the underlying code base prior to opening Pull Request:

cd MITunaX
./tests/covscripts/coverage_script.sh