Re: [Bloat] when does the CoDel part of fq_codel help in the real world?

[Mario Hock <mario.hock@kit.edu>]

Am 30.11.18 um 12:04 schrieb Jonathan Morton:
>> On 30 Nov, 2018, at 12:32 pm, Luca Muscariello <luca.muscariello@gmail.com> wrote:
>>
>> Two last comments: one should always used fluid approximation with care, because they are approximations,
>> the real model is more complex. Nobody considers that the RTT varies during the connection lifetime and that ACK can be delayed.
>> So CWD increases in a non simple way.
> 
> If I can restate that in a more concrete way: the queue may not drain at a smooth, constant rate.  There are several real-world link technologies which exhibit this behaviour - wifi and DOCSIS come to mind, not to mention 3G/4G with variable signal strength.
> 
>   - Jonathan Morton

That's true. In this case the the BDP of the path changes alongside the 
bandwidth. Thus, it's pretty hard for a congestion control to have 1*BDP 
(distributed among all concurrent flows) in flight. I think we would 
have to work with something like an average sending rate and accept 
short term queues when throughput drops.

With LoLa we can also set a lower bound for the delay. If the delay is 
below that threshold, we assume the link to be not saturated. This is 
mainly to filter out noise. But maybe this could be tweaked to cope with 
such technologies.

As far as know, 3G/4G already does queuing per user. Thus, if a user 
produces some delay (e.g., to be able to use high bandwidths, when it 
suddenly becomes available), it only affects the user itself. Other user 
could still focus on lower delays.

Best, Mario