Server Traffic Prediction Using Machine Learning for Optical Circuit Switching Scheduling

Abstract:
This work presents the performance analysis of nonlinear autoregressive neural networks applied in server traffic prediction and dedicated to hybrid intra-data center networks. This technique offers a high potential for accurate forecasting of heavy traffic streams, which allows for an a priori scheduling and allocation of optical switching circuits for offloading of intensive flows in data center applications. Just as with the artificial data sets carrying the machine-generated traffic characteristics used in our previous work, the nonlinear autoregressive neural networks show a high performance of the prediction mechanism applied to real server traffic.