Machine Learning Assisted Optical Network Resource Scheduling in Data Center Networks

Parallel computing allows us to process incredible amounts of data in a timely manner by distributing the workload across multiple nodes and executing computation simultaneously. However, the performance of this parallelism usually suffers from network bottleneck. In optical switching enabled data center networks (DCNs), to satisfy the complex and time-varying bandwidth demands from the parallel computing, it is critical to fully exploit the flexibility of optical networks and meanwhile reasonably schedule the optical resources. Considering that the traffic flows generated by different applications in DCNs usually exhibit different statistical or correlative features, it is promising to schedule the optical resources with the assistance of machine learning. In this paper, we introduce a framework called intelligent optical resources scheduling system, and discuss how this framework can assist resource scheduling based on machine learning approaches. We also present our recent simulation results to verify the performance of the framework.