Deep-RL: Deep Reinforcement Learning for Marking-Aware via per-Port in Data Centers

In this paper, we propose Deep-RL—a marking decision with Deep Reinforcement Learning (DRL) via per-port for solving the erroneously marking problems in multi-queue multi service scenarios of Data Center Networks (DCNs). We formulate the statement as a DRL problem and use Deep Neural Network (DNN) to achieve the best possible policy for the agent. In this way we can model the complex DCNs in order to obtain the optimal threshold in the output port when the marked packets from queue buffers are not a concern. Unlike prior research that focused on mathematical models or used machine learning in DCN, Deep-RL is a novel DRL based method, which optimizes the real value of threshold with continues action space. Thus, this fact makes our work incomparable with previous research. To the best of our knowledge, we are the first to discuss the problem with DRL and DNN. Simulation results demonstrate that Deep-RL utilizes the buffer capacity at exactly 30% and achieves near optimal flow completion time.