Toggle navigation
HY Blog
Home
About
Tags
HY Blog
New post every day (with probability 0.03).
深度增强学习在数据中心网络中应用
论文汇总
深度增强学习在数据中心网络中应用 AuTO:Scaling Deep Reinforcement Learning for Datacenter-Scale Automatic Traffic Optimization (Sigcomm 2018) Neural Adaptive Video Streaming with Pensieve (SIGCOMM 2017) Resource...
Posted by Yiran on March 27, 2019
Useful links
Research and Paper Writing, Miscellaneous
Research and Paper Writing: How to Do Great Research (Grad school survival advice from Nick Feamster and Alex Gray) How to Read Writing Technical Articles Common Bugs in Writing THE PH.D...
Posted by Yiran on March 27, 2019
Timely SIGCOMM 2015
RTT-based Congestion Control
核心思想 第一个RTT-based congestion control在数据中心的应用,认为RTT是个很好的拥塞信号。 拥塞信号的信息量 传统TCP的ECN ————> 超过一定阈值才标记,只有0 和 1的变化 DCTCP的ECN ————> 更进了一步,将单bit转化为一个RTT内的多bit 但是ECN的方式对多优先级不general:低优先级不被标...
Posted by Yiran on March 27, 2019
Thoughts of Programmable Switches SIGCOMM 2018 CCR
The Role of Programmable Switches
核心思想 在可编程交换机上实现的有两类:应用功能(Load Balancer比如SilkRoad、Load partition比如NetCache) 和 网络功能 (Network traffic balancing、Network Telemetry、Packet scheduling、Congestion Control) 其实几乎所有的应用的offload都可以通过serv...
Posted by Yiran on March 27, 2019
SIMON NSDI 2019
Network Tomography技术在数据中心测量的应用
核心思想 用网卡收集到的信息重构网络中队列信息、链路利用率等(Network Tomography技术) Measurement 和 monitoring 的三大挑战 准确性、可扩展性、速度 已有工作分类 分为 switch-based 和 edge-based. switch-based: 可以收集到很细粒度的信息,但是代价高,需要把多个交换机信息拼接在一起、大量带宽去传递...
Posted by Yiran on March 27, 2019
Pingmesh SIGCOMM 2015
Data Center Network Latency Measurement and Analysis
核心思想 构建一个大规模的数据中心latency测量和分析系统(微软) 网络运维面临的挑战 确定是否是网络的问题、定义和追踪SLA、network troubleshooting Design Pingmesh Controller: 负责产生pinglist.为了减少probe数目, 设计了多重complete gragh. 分为inter-dc和intra-dc. in...
Posted by Yiran on March 27, 2019
Flux NSDI 2019
Flow size prediction
核心思想 机器学习算法预测流的size, 除此之外的亮点:探讨了对流大小是不是知道的越多越好? 获取流大小 方法总结 应用提供 局限:有的应用一开始不知道大小;需要改应用API,对public cloud不适用 Flow aging (LAS PIAS) 局限:不适用于需要知道流确切大小的调度场景(fastpass、sincron...
Posted by Yiran on March 27, 2019
DETER NSDI 2019
Deterministic TCP Replay
核心思想 使得operator能够反复、确定地重放tcp的trace,从而使用各种工具,作出问题诊断。 Oh,,TCP能重放的话 就能量化不同bottleneck了!并且不需要改应用! 主要挑战 butterfly effect(蝴蝶效应)—a small timing variation causes a chain reaction between TCP and the netw...
Posted by Yiran on March 27, 2019
← Newer Posts
FEATURED TAGS
Measurement in Datacenter
TCP Performance Diagnosis
Flow Scheduling
Machine Learning
Network Troubleshooting
Programmable Switches
Congestion Control
Transport in Datacenter
ABOUT ME
Goals determine what you going to be!
✉️ yeholdon@gmail.com
FRIENDS
WY
简书·BY
Apple
Apple Developer