TEL: Low-Latency Failover Traffic Engineering in Data Plane
share
Modern network applications demand low-latency traffic engineering in the presence of network failure while preserving the quality of service constraints like delay, and capacity. The control plane reactions to the failure can be slow compared to the data plane while supporting traffic demands for highly sensitive applications. The control plane interaction requires to find an alternative path for the failed one in the legacy approaches. In this paper, we formulate failover traffic engineering as a max-min fair allocation problem that maximizes the number of flows while minimizing their costs. We also present TEL, a system with a linear algorithm, that uses the idea of backup paths to avoid the control plane interaction to compute new paths. We use a reinforcement learning-based algorithm to explore paths in the network. In particular, our solution performs traffic engineering in the data plane. We implement our approach in P4 and evaluate it on two real-world topologies, namely, Goodnet and AttMpls. The simulation results confirm that TEL has significant throughput improvement and lower flow completion time compared to Open Shortest Path First (OSPF). Finally, we state the applicability of TEL in the different modern network applications.
READ FULL TEXT