- Definition: Technology to control and manage networks through software.
- Features: Separate the Control Plane and Data Plane of the network.
- Existing Network: Control planes are distributed to each network equipment.
- SDN: Provides an integrated control plane.
-
Network Simulator: Minnet
-
SDN Controller: RYU
-
Deep Learning Library: PyTorch
-
- Conditions: Benchmark execution based on 'simple_switch'.
- Principle of operation:
- Save the source MAC address to the table.
- If the destination MAC address is in the learning table, it is forwarded to that port.
- Flooding packets if not learned.
- Performance:
-
평균 Throughput: 3312.12 responses/s
-
Standard deviation: 198.10 (smaller stable)
-
- Conditions: Same test environment (128 switches, 3 hours running).
- Using optimization techniques such as GPU porting, tiling, caching, etc.
- Performance:
-
평균 Throughput: 8949.96 responses/s
-
Standard deviation: 249.12 (smaller stable)
-
-
- The resulting data can be visualized as a graph to confirm performance improvement (using e.g., Cbench).
- Deep learning optimization of SDN significantly improves throughput over existing methods.
- Proves the potential of deep learning technology in intelligent and performance optimization of network management.