Re: [aqm] think once to mark, think twice to drop: draft-ietf-aqm-ecn-benefits-02

[David Lang <[e-mail]>]

On Fri, 27 Mar 2015, De Schepper, Koen (Koen) wrote:

> Hi all,
>
> Agreed that there implementations should not be described. Also agree that we should describe all the benefits of using ECN, so I propose to reformulate something as follows:
>
> ECN allows to introduce and deploy state of the art ECN based congestion controllers,
> and to guarantee throughput fairness by applying in the network an appropriate
> relation between marking and dropping.
>
> The extra benefits of supporting these improved ECN based congestion controllers are:
>
> - low latency AND high throughput (compared to low latency OR high throughput for drop
> based congestion controllers)

I think that you are overstating things when you say that without ECN you are 
forced to choose between low latency OR high throughput. that doesn't match what 
people are reporting when they use simple fq_codel without ECN

> - reduced variability in flow fairness between competing flows because of the 
> high marking rate

This is plausable, but how unfair do things get in practice? Especially if you 
have a large number of flows so that you are statistically unlikely to keep 
picking on the same flow.

> - scalability to higher throughputs because the marking rate is independent 
> from the throughput

The drop rate is also independent of the throughput, so I don't see how this 
works unless you are doing statistical dropping.

If you take the approach that there is a probability P that you should drop/mark 
a packet, these statements make a lot of sense. But if you are doing something 
like fq_codel that is based on incoming packets, not probability checks, they 
make less sense.

Yes, you can analyse the results by saying that there was a probability P of a 
packet being dropped, but that's not what's being done under the covers.

> Disadvantage is that congestion controllers that are optimized for drop 
> signals cannot use ECN without experiencing high throughput unfairness.

This is being disputed heavily by others here who are saying that adding ECN 
cannot possibly cause unfairness (in either direction)

David Lang