Network Papers in NSDI 2026¶
Tip
updated on 2026-01-21
(1) MirrorNet: High-fidelity and Scalable Network Emulation for Software-defined WAN
(2) Harvesting Spare CPU Resources in Container Systems [HarvestContainers]
(3) HyperEdge: An Edge CDN Infrastructure for Cost Efficient Video Streaming
(4) A Fast Solver-Free Algorithm for Traffic Engineering in Large-Scale Data Center Network [SSDO]
(5) SLATE: Service Layer Traffic Engineering
(6) Law: Towards Consistent Low Latency in 802.11 Home Networks [LAW]
(7) QCON: Seamless QoE-Aware 5G Streaming via Multi-Connectivity
(8) Managing Congestion Control Heterogeneity on the Internet with Approximate Performance Isolation [SANTA]
(9) CascadeNet: Generating Network Traffic with High-Fidelity Temporal Patterns
(10) Matryoshka: Realizing Hyperscale Data Center Network Design for the AI Era
(11) CCEval: Accurately and Confidently Evaluating Performance Metrics of Congestion Control Algorithms for Datacenter Networks
(12) HEDGE: Traffic Engineering with Probabilistic Link Capacities
Summary
NSDI26:
| 论文 | 核心 | 推荐系数 | 亮点与评价 |
|---|---|---|---|
| MirrorNet | 腾讯WAN生产系统的数字孪生 | 0 | 纯工业界水文, 千万别看 |
| HarvestContainers | 给延迟敏感容器始终留几个真正空着的核作为 Buffer, 让它的线程被唤醒时像独立运行一样不排队; 其余空闲核借给吞吐型容器, 并用 Kernel 忙轮询加 Two-Phase Affinity加快调用速度 | 5 | 深度好文, 后面做纯 System Work 要回头看. 跟 MobiSys 那个 Ringmaster 都要仔细看 |
| LAW | 新型 WiFi 802.11 链路层架构 | 1 | 感觉不是凭质量中NSDI的... 主要在queue层级上进行了优化(MAC/Driver/HW), 有点像AQM |
| QCON | |||
| SANTA | 采用"近似性能隔离"的 AQM, 通过 "同性同队" + "异性shuffle" 实现 | 5 | 非常好文章, 在不质疑eval的前提下, 确实有很多可学之处 (1) 提出了 delay-loss-thpt "三位一体"坐标轴 (2) 核心洞察: 具有相似"吞吐量-延迟"偏好的CC, 在共享同一队列时, 能够友好共存并达成它们各自期望的性能目标 ("同类算法友好共存, 异构算法相互伤害") (3) 互联网中约 90% 的流是短流("mice flow") (4) 解法直观: 少量queue. 采用shuffle机制, 尽全力保证"同类同queue", 实现"三位"的近似性能隔离 (5) 分类整理了 E2E CCAs, Network-Assisted Approaches, Fair Queue, AQMs. 还有 "Beyond bandwidth fairness", 如: FCT, NUM, TCP-friendly |
| CascadeNet | 基于 GAN 实现高保真 Trace 生成 | 1 | 出发点: 不让 GAN 去学逐包的时间戳，而是把每条流压缩成定长的聚合时间序列来学，最后把逐包 trace 还原出来 锐评: 非常黑魔法. 完全不值得学习, 以后有需要引用灌灌水就行 |
| Matryoshka | 生产级别的 DCN 设计系统，系统地将 High-level DCN 设计意图编译为有效的交换机配置 | 0 | 纯工业界水文, 千万别看 |
| CCEval | 有点像是 CCA Arena. 搭建"DCN CCAs测评"平台 | 2 | 是个不错的CCA入门综述, 但客观来说不认为能过 NSDI 的 bar (1) CCA Evaluation 三大步骤 (2) 常见 DCN CCAs / Emu / Simu 积累 (3) CCAs 常见分类方式 |