Hierarchical Two Dimensional Queuing: A Scalable Approach for Traffic Shaping using Software Defined Networking

Maintaining quality of service (QoS) at network cores is particularly challenging because of the dynamic service requirements from long (elephant) and short (mice) flows. Traditional integrated and differentiated services have high overhead in this scenario because they work on packet level classification rather than on flow level classification, and also it is difficult to estimate the QoS requirements a priori. In this paper, we explore a software controlled architecture to implement a hierarchical control and management framework for QoS provisioning at network cores. We show that both the service differentiation and fairness are required to maintain QoS in such scenario. Accordingly in this paper, we propose an optimization followed by a two dimensional queue management policy, called Hierarchical Two Dimension Queuing (H2DQ), over a software controlled differentiated service architecture to ensure both the service requirements and fairness for the elephant as well as mice flows. We have implemented H2DQ over a testbed using virtualized architecture, and show that H2DQ can significantly improve the performance in terms of QoS and fairness performance, while having less control overhead.