[Codel] self tuning codel

[Jonathan Morton]

> On 19 Jul, 2018, at 10:04 pm, Kathleen Nichols <nichols at pollere.com> wrote:
> 
>> Thus, we let a specific AQM router serve
>> as a centralized DR in the system. All routers send packet
>> dropping probability updates to DR. DR then updates its list
>> of packet dropping probabilities and sends it to all routers
> 
> Does this seem possible to anyone? There's an inherent assumption here
> that "dropping probabilities" are constant over a period of time that is
> long compared to the time to do the above. In what I've seen of traffic
> and AQMs this seems unlikely but I'm certainly not as well-informed on
> this as some people.

Based on the above, the approach seems completely wrong-headed to me.  AQM should not be centralised, but must act according to local conditions.  Certainly such a system would be impossible to deploy in a heterogeneous environment like the Internet.

>> every AQM router
>> needs to know all the packet dropping probabilities at the
>> bottleneck links in order to compute the window sizes of the
>> flows from (5).

As a first approximation, you can assume that if you are the bottleneck, nobody else on the same path is, and therefore their drop probabilities are all zero.  This assumes only that you're dealing with traffic that's responsive to congestion signals, as TCP is.  Transient conditions may exist where the assumption isn't entirely correct, but these are *only* transient.

The situation for UDP may of course be different.

 - Jonathan Morton