This repository provides a modular simulation framework for dynamic resource allocation in 5G-enabled smart city infrastructures. It leverages Multi-Agent Reinforcement Learning (MARL) to allocate computing and network resources efficiently across heterogeneous service types like eMBB, URLLC, and mMTC.
The simulation framework includes:
- A substrate network representing physical infrastructure.
- Virtual network hierarchical slices for samrt city applications.
- Smart city service request generation.
- Multi-agent Deep Q-Learning (DQN) based resource allocation.
It is designed for research and experimentation in intelligent network management, network slicing, and edge computing.
| File | Description |
|---|---|
main.py |
Entry point that initializes and runs the simulation. |
ReSA-MARL.py |
Core MARL environment and training loop. |
SubstrateNetworkManager.py |
Physical infrastructure and substrate topology. |
VirtualNetworkManager.py |
Virtual slices, CNFs/SCFs, and hierarchical smart city services. |
RequestManager.py |
Smart city service request modeling and traffic profiles. |
ResourcesManagement.py |
Assumptions for CPU, RAM, storage, PRBs, and bandwidth needs. |
AllocationManager.py |
(or Placement Manager) Resource mapping logic from virtual to substrate nodes. |
Ensure you have Python 3.8+ installed.
pip install -r requirements.txtRun the full simulation using:
python main.pyThis will:
- Generate a substrate network and virtual slices.
- Simulate smart city request traffic.
- Train DQN agents to perform multi-agent resource allocation.
This framework is modular and fully customizable:
-
Modify hierarchical slices and virtual node setup in
VirtualNetworkManager.py -
Adjust the substrate network's topology and resource capacities in
SubstrateNetworkManager.py -
Simulate different service request types and traffic demands in
RequestManager.py -
Tune QoS-based resource requirement formulas (CPU, RAM, Storage, PRBs, Bandwidth) in
ResourcesManagement.py
This design allows you to simulate a wide range of 5G and smart city deployment scenarios.
- Trained models:
model_weights/ - Logs and events:
logs/ - Simulation results and resource mappings:
json_files/
- Real-time visualization of virtual-to-substrate mappings.
- Graph theory based arguments for configurable topologies and resource allocation patterns.
- Pre-trained models for benchmarking.
This project is licensed under the MIT License.
Pull requests are welcome. For major changes, please open an issue first to discuss your ideas.
F. Saleh, A. O. Fapojuwo, and D. Krishnamurthy, “eSlice: Elastic Inter-Slice Resource Allocation for Smart City Applications,” IEEE Transactions on Network and Service Management, Early Access, pp. 1–1, 2025, doi: 10.1109/TNSM.2025.3604352.