[Bloat] [Codel] The challenge

[Eric Dumazet]

On Thu, 2012-05-10 at 02:02 +0200, Steinar H. Gunderson wrote:
> On Thu, May 10, 2012 at 01:58:39AM +0200, Eric Dumazet wrote:
> > One solution is at the producer level (TCP stack of the sender in this
> > case) to properly pace the transmits in MSS units given the RTT and
> > cwnd, supposedly done with TCP Hybla.
> 
> You can also rate-limit individual flow, if you're happy with an upper
> bound on each flow (in my case: video streaming). I'm testing this these
> days, and without having any proof, I'm quite convinced it reduces
> bufferbloat problems significantly.

Of course, if you have a given throughput per flow, you can do that.

TCP stack has no clue about this an try to fill the pipes as fast as
possible.