Control channel isolation with time-series control traffic prediction in programmable network virtualization

A network hypervisor apparatus for providing a software defined networking (SDN)-based virtual network, the network hypervisor apparatus comprising:

• a data collector configured to collect control traffic data and network topology information for each virtual switch;
• a control traffic predictor configured to predict future control traffic based on the control traffic data and the network topology information;
• and
• a translator configured to translate a control message from a virtual switch to a physical switch based on a prediction result,
• wherein the control traffic predictor is configured to perform prediction using a prediction model based on a pretrained long short term memory (LSTM)-autoencoder model,
• wherein the prediction model comprises:
• an encoder including LSTM cells corresponding to the number of past sequence data, each LSTM cell generating a new cell state for the past sequence data, and a last LSTM cell generating a single piece of compressed data in an encoded state;
• and
• a decoder including LSTM cells and activation functions, receiving the single piece of compressed data and the past sequence data, and predicting future sequence data when each of the LSTM cells outputs predicted data.