Horizon: a QoS management framework for SDN-based data center networks

Abstract

Data center networks (DCNs), which serve as the infrastructural backbone of big data, have been receiving increasing attention recently. To improve the service quality of data centers, researchers have been working on congestion control, network monitoring, and performance optimization. However, most such works focus on user-centric service quality, which means that the quality of network service itself is not factored into the quality of service (QoS) problem. In this study, we illustrate the problem of data center operations and management as a new type of QoS that is the foundation of user-centric QoS implementation. Inspired by traditional works on network performance optimization, we define the quality of network service in a software-defined networking (SDN)-based DCN and develop a framework called Horizon as the architecture of our QoS solution. This framework comprises a Markov-process-based method to predict link popularity, and we use SDN technology to monitor network status. We implement the proposed method, and the experimental results indicate that Horizon can relieve congestion in DCNs to meet QoS requirements. The experimental results show that our approach has a similar performance to the optimal solution. When compared with the ECMP approach, our approach has a much lower latency. The results also show that the proposed approach is effective in terms of network congestion control.