Bulk Transfer Scheduling with Deadline in Best-Effort SD-WANs

Many cloud providers have multiple geo-distributed inter-connected datacenters around the globe. These datacenters are increasingly being inter-connected using software-defined WANs (SD-WANs), which extend the capabilities of SDN architecture to wide-area networks. While conventional MPLS tunneling has proven to be a practical approach for inter-connecting datacenters, such tunnels have a static nature and incur substantial maintenance costs. Given the centralized control and programmability of SDN, it is possible to utilize multiple Internet tunnels to provide a low-cost alternative to MPLS tunnels in SD-WANs. However, the best-effort nature of Internet tunnels means that they undergo capacity fluctuations throughout the day, making it difficult to provide any service guarantees such as completion time for inter-datacenter transmissions. In this paper, we consider the problem of scheduling bulk transfer requests with deadline in a best-effort SD-WAN. We propose an approximate scheduling algorithm called xBESD which utilizes tunnel capacity estimations to design a robust transfer schedule that maximizes a cloud provider's profit by transmitting bulk transfers before their deadlines. We analyze xBESD and show that it attains an approximation ratio that only depends on the number of overlapping requests that have the same profit to bandwidth ratio. Furthermore, we provide extensive simulation as well as realistic Mininet experimental results to assess the performance of xBESD in a variety of network scenarios. Our results show that xBESD improves the provider's profit by approximately 60% on average compared to other baseline scheduling methods, in addition to cutting down the Internet service provider costs.