In this project, we propose a reinforcement learning-based DVFS scheduling algorithm for energy-efficient real-time systems. The proposed algorithm is designed to minimize the energy consumption of real-time systems while satisfying the timing constraints of tasks. Actually, the RL agent tries to keep the correct trade-off between frequency and voltage (higher frequency and voltage lead to higher energy consumption but faster and better performance). The proposed algorithm is evaluated on a set of synthetic task sets (having specific meaningful characteristics) using a customized version of UUniFast.

To install the required packages, you can use the following command:

pip install -r requirements.txt

It is recommended to use a virtual environment to avoid conflicts with other packages. (like conda)

conda create -n myenv python=3.8
conda activate myenv

To test and visualize the task set generation (using the custom UUniFast algorithm), you can use the following command:

python3 test_task_generation.py

and the results (charactersitics of the generated task sets) will be printed and the plots will be saved in the root folder.

To train the RL agent, you can use the following command:

python3 train_and_evaluate_agent.py

The results (including characteristics) will be printed and the plots will be saved in the results/ folder.

Also if you want to train your agent on pre-specified scenarios (including different utilization values and number of cores), you can use the following command:

python3 run_multiple_scenarios.py

The results (including characteristics) will be printed and the plots will be saved in the results/ folder.

You can see a few samples of the plots, below:

Also, the scheduled tasks and their details will be saved as a .csv file in the same directory.

We also propose a hybrid scheduling algorithm (hard-deadline periodic + soft-deadline aperiodic). To test the hybrid scheduling algorithm, you can use the following command:

python3 run_hybrid_scheduling.py

Which will result in logging different metrics and plots like below:

also if you want to see a sample generated aperiodic set of tasks, run the following command:

python3 test_aperiodic.py

Which will result in a plot like this: