Computer Networks · 免费预览

拥塞避免

与其等待缓冲区溢出,避免方案会预测拥塞并提前发出信号:DECbit 让路由器在平均队列长度超过 1 时设置拥塞位,源端则根据多数数据包到达时是否被标记来减半或增大窗口;随机早期检测计算加权平均队列长度,并在两个阈值之间以递增的概率丢弃到达的数据包,从而用少量早期丢包防止后期大量丢包;显式拥塞通知将信号携带在头部,而不是通过丢弃数据包;基于源端的方案则直接测量延迟,Vegas 将期望速率与实际速率相对于阈值进行比较,BBR 对瓶颈带宽和最小往返时间建模并按该模型调整发送节奏,DCTCP 在数据中心采集显式标记,按测得的拥塞比例缩减窗口。

1. Setting the Bit

Setting the Bit

2. Dropping Early, On Purpose

Dropping Early, On Purpose

3. Marks, Not Corpses

Marks, Not Corpses

4. Listening to Delay Instead

Listening to Delay Instead

5. The one-bit warning

Under DECbit, when does a router set the congestion bit in a packet's header?

6. Why drop on purpose

What is the purpose of RED's dropping packets before the buffer is full?

7. The averaging formula

RED computes its average queue length as AvgLen=(1−w)⋅AvgLen+w⋅SampleLen\text{AvgLen} = (1-w)\cdot\text{AvgLen} + w\cdot\text{SampleLen} with small ww. Why average at all instead of using the instantaneous queue length?

8. Marking beats dropping

Why is explicit congestion notification preferred over RED-style dropping where both endpoints support it?

9. Vegas's arithmetic

A Vegas flow has a window of 8 packets, a BaseRTT of 100 ms, and measures an actual rate of 60 packets per second. What does the flow conclude?

10. DCTCP's gentleness

What makes DCTCP's window adjustment gentler than classic halving on loss?

11. The predicted event

Congestion avoidance schemes try to predict and signal what event before it happens?

12. The probability curve

In RED, between the minimum and maximum thresholds, the drop probability does what as average queue length grows?

13. The yardstick

In TCP Vegas, what is BaseRTT, the round-trip time used as the denominator of the expected rate?

14. The model pair

BBR models a path using which two measured quantities?

进入交互式练习

免费注册 · 即时批改 · 自动安排间隔复习。

免费开始这门课