TetriX: A Decentralized and Resilient Multi-Agent Scheduler for HPC Clusters

TetriX is an advanced task scheduler designed for high-performance computing (HPC) clusters. Utilizing a decentralized and resilient approach, TetriX employs a network of agents for resource management, aiming for near-optimal outcomes and surpassing traditional human heuristics.

Features (Under Development)

• Efficient Scheduling: Advanced algorithms using multi-agent reinforcement learning for near-optimal task allocation.
• Decentralized Architecture: Even distribution of tasks for enhanced robustness.
• Resilience and Reliability: Designed to maintain functionality under various scenarios.
• Dynamic Adaptability: Real-time adjustments to workload changes.
• Scalability: Effective for both small and large HPC clusters.

File Structures

tetrix/
│
├── envs/                # Gym environments specific to TetriX
│   ├── __init__.py      # Makes envs a Python module
│   └── ...              # Environment implementation files
│
├── agents/              # Agents interacting in the environments
│   ├── __init__.py      # Makes agents a Python module
│   ├── dqn_agent.py     # Example: DQN agent implementation
│   └── ...              # Other agent types
│
├── train/               # Training scripts for the agents
│   ├── train_dqn.py     # Script to train DQN agent
│   └── ...              # Other training scripts
│
├── models/              # PyTorch models (Neural Networks)
│   ├── __init__.py      # Makes models a Python module
│   ├── dqn_model.py     # Example: DQN network architecture
│   └── ...              # Other model architectures
│
├── utils/               # Utility code
│   ├── __init__.py      # Makes utils a Python module
│   └── ...              # Helper functions and classes
│
├── requirements.txt     # Python dependencies, including PyTorch
├── README.md            # Project overview and setup instructions
└── LICENSE              # License information