Traffic Prediction for Inter-Data Center Cross-Stratum Optimization Problems

In this paper, we consider resource allocation in data networks and evaluate the performance of various approaches using the Cross-Stratum Optimization architecture for a providercentric use case. We describe many approaches used to provision various requests related to a data center operations. We consider three simple approaches and compare them with two additional ones: the hybrid method that simultaneously considers cost, distance, and utilization; and traffic prediction methods based on Monte Carlo Tree Search that employs machine learning techniques. In evaluations, we rely on data center models and pricing structure provided by Amazon Web Services. Results indicate that using approaches that jointly optimize two strata improve network performance. Finally, the use of machine learning techniques enables network and data center operators to more efficiently utilize resources.