First-Slot scheduling with wavelength conversion for distributed file transfers

Abstract

As more flexible and self-adaptive network connections are expected between cloud users and service vendors, the requests for more sophisticated bandwidth scheduling schemes are emerging. In this paper, we conduct a thorough study on the a specific type of bandwidth reservation: First-Slot scheduling in this scenario. We introduce a model for the optical networks with sparse wavelength conversion, which are now widely served as dedicated networks between the users and the cloud. We evaluate the performances of EBF and KDP algorithms under various topologies and workloads. Our results show that increasing the fraction of nodes with wavelength converters reduces the blocking probability, but have only marginal effects on requests' average start time. However, both metrics are heavily affected by network topology and traffic load. We also observe that accepting requests greedily, like EBF does, yields better performance when the workload is low, but the more conservative approach, KDP, is superior when the workload is high.