Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/bruhathisp/nocsimulation

Dynamic Optimization of Network-on-Chip
https://github.com/bruhathisp/nocsimulation

Last synced: about 2 months ago
JSON representation

Dynamic Optimization of Network-on-Chip

Awesome Lists containing this project

README 

        
# NoCSimulation

Dynamic Optimization of Network-on-Chip



# NoCTune Project README



## Overview



NoCTune is an advanced simulation framework developed to optimize Network-on-Chip (NoC) systems using Reinforcement Learning (RL). This project integrates an OMNeT++ simulation environment with a custom-developed Actor-Critic model in PyTorch to dynamically adjust NoC configurations based on real-time performance metrics.



## Goals



The primary goal of NoCTune is to design a NoC that is area and power efficient while achieving optimal performance in terms of latency and bandwidth. The simulation considers various traffic patterns and workload scenarios on a SoC that includes CPU and IO peripherals accessing system memory through a weighted arbiter.



1. (Write pseudocode to measure average latency and bandwidth using the simulator provided monitor output (as shown in Table 1.0). The pseudocode needs to be efficient and robust.)[https://github.com/bruhathisp/NoCSimulation/goals/Goal1/README.md]

2. (Using Reinforced Learning to arrive at the optimal parameters listed above. Put together a design document. Describe the RL framework consisting of states/behaviors, actions and rewards. You need to also advise which RL algorithm (ex: DQN, SARSA, Actor-Critic, etc.) is best suited for this problem statement. You need not develop or write the  RL algorithm. ) [https://github.com/bruhathisp/NoCSimulation/goals/Goal2/README.md]



## [Demo Video](https://www.youtube.com/watch?v=8IhDGbxIHnk)



## Features



- **Custom NoC Simulation Environment**: Simulates complex interactions within a SoC, including data flows between CPU, IO peripherals, and system memory.

- **Dynamic Routing and Buffer Management**: Uses real-time performance data to adjust routing paths and buffer sizes.

- **Actor-Critic Reinforcement Learning Model**: Optimizes system configurations for latency, bandwidth, and power consumption.

- **Performance Monitoring**: Tracks and adjusts system performance based on predefined thresholds for latency, bandwidth, buffer occupancy, and power consumption.



## Installation



### Prerequisites



- Python 3.6+

- PyTorch 1.7+

- Gym 0.17+

- OMNeT++ 6.0.3



### Setup



1. **Clone the repository**:

   ```bash

   git clone https://github.com/your-repository/NoCTune.git

   cd NoCTune

   ```



2. **Install Python dependencies**:

   ```bash

   pip install torch gym numpy

   ```



3. **Set up OMNeT++** (Assuming OMNeT++ is already installed and configured):

   - Configure the OMNeT++ environment to integrate with the Python scripts.



4. **Build the simulation**:

   ```bash

   cd src

   make makefiles

   make

   ```



## Usage



### Running the Simulation



1. Installation Guide of Omnetpp - https://doc.omnetpp.org/omnetpp/InstallGuide.pdf 



2. **Run the reinforcement learning model** 



### Configuration



Adjust the `omnetpp.ini` file to set different simulation parameters like traffic patterns, buffer sizes, and arbitration weights.



## Custom Environment Setup



The `NoCEnv` class in `model.py` interfaces with the NoC simulator. It manages the complexities of system states, actions, and reward calculations.



## Architecture



### Modules



- **network/**: Contains OMNeT++ modules like CPU, IO, Router, and Arbitrator.

- **util/**: Includes utilities like the TestBench module for testing.

- **model/**: Contains the Actor-Critic model implemented in PyTorch.

- **messages/**: Definitions for custom message types used in the simulation.



### Reinforcement Learning Model



The Actor-Critic model uses the following key components:

- **Actor Loss**: Encourages the policy to take actions that lead to higher returns.

- **Critic Loss**: Penalizes inaccurate predictions of future returns.

- **Entropy Bonus**: Encourages exploration by preventing premature convergence.



## Contributing



Contributions are welcome. Please fork the repository and submit pull requests with your enhancements.



## License



Distributed under the MIT License. See `LICENSE` for more information.



## Contact



Your Name - [email protected]



Project Link: https://github.com/bruhathisp/NoCSimulation/