FLAIR: Accelerating Reads with Consistency Aware Network Routing

on YouTube ↗20m

" summary

Proposes FLAIR, a network-side pipeline implemented in P4 on a Barefoot Tofino switch that tracks client requests and replies to mark key ranges stable and forward reads to consistent followers, preserving linearizability. Implements FlairKV on Raft, uses 12 registers and ~30 tables, and handles concurrent writes, packet reordering, and switch failures. Evaluation shows up to 42% higher throughput and large latency reductions versus leader-only and lease-based approaches (uploader reports 35–97% lower latency).

AI-written summary · about

" chapters (10)

1. 0:00Start
2. 3:52Evaluation Summary
3. 4:21Programmable Switches Overview
4. 5:53Programmable Switches Challenges
5. 6:32FLAIR Design
6. 7:25FLAIR's Object Stability Array
7. 12:10Implementation
8. 13:38Throughput
9. 14:54Read Latency
10. 16:45Conclusion

" related talks

• 16mFLB: Fine-grained Load Balancing for Lossless Datacenter NetworksKai Chen, Wenxue Li, Jiawei Huang, Jianxin Wang, Jinbin Hu, Xiangzhou Liu, Junfeng Wang, Bowen Liu, Ping Yin · USENIX ATC '25
• 19mFarReach: Write-back Caching in Programmable SwitchesPatrick P. C. Lee, Qun Huang, Siyuan Sheng, Huancheng Puyang, Lu Tang · USENIX ATC '23
• 15mFlor: An Open High Performance RDMA Framework Over Heterogeneous RNICsChao Han, Dennis Cai, Xiaoliang Wang, Bo Li, Guihai Chen, Qiang Li, Chen Tian, Peng Zhang, Hongqiang Harry Liu, Qiao Xiang, Jianbo Dong, Fei Feng, Jiaji Zhu, Yixiao Gao, Lingbo Tang, Yaohui Wu, Shaozong Liu, Lei Yan, Zhongjie Wu, Zheng Cao, Jinbo Wu, Rui Miao, Yanfang Le, Haonan Qiu, Derui Liu, Zhiwu Wu, Weijie Li · OSDI 2023
• 23mBuilding a network stack for optimal throughput / low-latency trade-offsNico Kruber · Flink Forward Berlin 2017
• 16mRingLeader: Efficiently Offloading Intra-Server Orchestration to NICsAditya Akella, Jiaxin Lin, Tarannum Khan, Hassan M. G. Wassel, Brent E. Stephens, Adney Cardoza, Yeonju Ro · NSDI 2023
• 18mTiara: A Scalable and Efficient Hardware Acceleration Architecture for Stateful Layer-4Kai Chen, Teng Zhang, Chaoliang Zeng, Zilong Wang, Layong Luo, Tao Feng, Xiongfei Geng, Chuanxiong Guo, Luyang Li, Wenchen Han, Nan Chen, Lebing Wan, Lichao Liu, Zhipeng Ding, Feng Ning · 19th USENIX Symposium on Networked Systems Design and Implementation (NSDI '22)
• 14mEmpower Programmable Pipeline for Advanced Stateful Packet ProcessingYong Feng, Bin Liu, Yang Xu, Zhikang Chen, Haoyu Song, Chuwen Zhang, Yinchao Zhang, Hanyi Zhou, Ruoyu Sun, Wenkuo Dong, Peng Lu, Shuxin Liu · 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24)
• 16mPRED: Performance-oriented Random Early Detection for Consistently Stable Performance...Tong Li, Zhuotao Liu, Ke Xu, Xiangyu Gao, Kun Tan, Xinle Du, Guangmeng Zhou, Hanlin Huang, Mowei Wang · NSDI '25