Re: [tsvwg] feedback and thoughts L4S / SCE

[Jonathan Morton <[e-mail]>]

> On 22 Nov, 2020, at 4:46 pm, Sebastian Moeller <[e-mail]> wrote:
> 
>> I'd be surprised if it was enabled on more than 1% of residential connections.
> 
> 	[SM] This is a good question. Anybody here can offer an sufficiently robust estimate of the total number of residential connections? 

Well, for a start, take basically 100% of wireline free.fr subscribers, the largest wireline ISP in France, who for a number of years now has deployed fq_codel at its bottlenecks (including in CPE) as a matter of policy.  Is France 1% of the world?  No, but it's a darn good start, and free.fr is clearly not alone according to Jake Holland's data, which shows significant ECN activity on about half a dozen ASNs and an increasing trend overall.  I've also heard from reliable sources that Telenor is rolling out a similar deployment in Norway, and Preseem has helped to deploy many WISP networks that use fq_codel actively.

Let's be precise about this.  This does not imply that a higher than average proportion of free.fr, Telenor or WISP subscribers have ECN fully enabled in their endpoint hosts (as opposed to passively enabled as most servers do).  That means we can only actually see ECN activity on some small fraction of the deployment.  The corollary is that the deployment of AQM at bottlenecks is probably much wider than ECN data alone can show.  Take the latter as a *very* conservative lower bound.

Importantly, RFC-3168 AQMs provide the same rate of congestion signalling to Not-ECT traffic, via packet drops, as they do to ECT traffic using CE marks.  So all the interactions between TCP Prague and conventional CC algorithms apply just the same, and we have fresh data proving that.

Pete and I are preparing a presentation of concrete data to explain this situation more clearly.

 - Jonathan Morton