Switching Fabric Control with AI and ML Support

In this paper, I present a new way to improve the statistics of control algorithms for blocking log2N switching fabrics. The specific graph form of the internal state representation allows creating tags for machine learning and artificial intelligence systems for decision analysis. The proposed system of graphs represents the internal state of fabric, configuration of connections, and the relation between following states is also modeled and prepared for further analysis. Two types of analysis are presented: static - where simulations create connections, allowing for the preparation of more effective algorithms, and real-life - where real traffic is monitored, and decisions are made based on learned patterns, their weights, and the quality of results. Both are implemented on Virtex V - the main hardware FPGA chip of NetFPGA Card.