Journal of Cloud Computing: Advances, Systems and Applications (Oct 2023)

sRetor: a semi-centralized regular topology routing scheme for data center networking

  • Zequn Jia,
  • Qiang Liu,
  • Yantao Sun

DOI
https://doi.org/10.1186/s13677-023-00521-8
Journal volume & issue
Vol. 12, no. 1
pp. 1 – 15

Abstract

Read online

Abstract The performance of the data center network is critical for lowering costs and increasing efficiency. The software-defined networks (SDN) technique has been adopted in data center networks due to the recent emergence of advanced network control and flexibility demand. However, the rapid growth of data centers increases the complexity of control and management processes. With the rapid adoption of SDN, the following critical challenges arise in large-scale data center networks: 1) extra packet delay on the separated control plane and 2) controller bottleneck in large-scale topology. We propose sRetor in this paper, a topology-description-language-based routing approach for regular data center networks that leverages data center networks’ regularity. sRetor aims to reduce the packet waiting time and controller workload in software-defined data center networking. We propose to move partial forwarding decision-making from the controller to switches to eliminate unnecessary control plane delay and reduce controller workload. Therefore the sRetor controller is only responsible for troubleshooting complicated failures and on-demand traffic scheduling. Our numerical and experimental results show that sRetor reduces the flow start time by over 68% and the fail-over time by over 84%.

Keywords