HPLB: High precision load balancing based on in-band network telemetry in data center networks

Weimin Gao, Jiawei Huang, Ning Jiang, Zhaoyi Li, Shaojun Zou, Zhidong He, Jian-Xin Wang

Abstract: In modern data centers, the congestion-aware load balancing makes rerouting decisions according to the traffic load. However, it is difficult to accurately obtain network load status using limited congestion information. Recently, In-band Network Telemetry (INT) has been embedded in the latest merchant silicones to support information collection of realtime traffic state from network equipments. In this paper, we present High Precision Load Balancing (HPLB), which is a load balancing method based on in-band network telemetry. With INT, HPLB collects the load information of each node on the transmission path and feeds it back to the near-source switch, which then selects the forwarding path according to the leaf-to-leaf congestion information. The test results of large-scale evaluation show that, compared with ECMP, RPS, DRILL and CONGA, HPLB reduces the flow completion times by $$\sim$$ ∼ 6-71% and improves link utilization by $$\sim$$ ∼ 13-80%.