linux内核流控（traffic control）相关概念

这里是linux内核流控相关的基本概念，具体的代码请参考kernel/net/sched/目录下的相关代码

六个基本操作

1. Shaping
2. Scheduling：(re)arranging queued packets.
3. Classifying：separating packets/flows.
4. Policing: Matching a packet/flow to a rate.
5. Dropping: discarding a packet/flow.
6. Marking: Modifying the packet itself.

实现机制

1. Queues.
2. Tokens.
3. Buckets.

Linux 对象

1. qdisc: the queuing discpline.
2. class: a flexible internal queue.
3. filter: for classifying and policing.
4. classifier: identifying flow/packets.
5. policer: limiting matching flows.

基本原则

1. For shaping traffic on a link, always be the bottleneck.
2. You can only shape trasmiited traffic.
3. Every interface must have a qdisc.
4. Any newly created class contains a FIFO.
5. A classful qdisc with no children classes only consumes CPU.
6. Classes attached directly to the root qdisc can simulate vritual circuits.
7. A filter can be attached to classes or a classful qdisc.

Classless qdisc

1. FIFI: first-in, first out.
2. SFQ: stochastic fair queue.
3. TBF: token bucket filter.
4. GRED: generic random early drop.

Classful qdisc

1. CBQ: Class based queue.
2. HTB: hierarchical token bucket.
3. PRIO: prioritizing.
4. WRR: Weighted round robin.

SFQ

TBF