Re: types of traffic in tcp congest control

[Alan Cox <[e-mail]>]

> Well, if only the mice can still congest the link, which shows that the
> network resources are under severe shortage. The only way to solve under
> this situation is hardware solution. But I think even under this situation

I disagree. You can start by collapsing events together when there is
congestion. The fact that userspace can't easily ask "if I send this will
it go out right about now" is a socket API lack.

> restrict the elephants will also do benefit to alleviate the stage of
> congestion.

The bigger problem with your streaming high datarate connections is capture
effect. That's *very* non trivial to handle without doing expensive things like 
connection tracking, or per ToS queues and praying someone used the ToS bits
(or port based hacks)

Alan