Experimental assessment of traffic prediction assisted data center network reconfiguration method

We experimentally demonstrate a traffic prediction assisted network reconfiguration method (TPANR) for data center networks based on deep reinforcement learning (DRL). Traffic prediction model performs the lowest MSE of 2.64E-4. Exploiting one-step ahead traffic prediction and DRL-based automatic network reconfiguration, TPANR achieves 17.3% latency improvement.