Re: [tsvwg] David Black (individual) on safety of L4S for the Internet

[Sebastian Moeller <[e-mail]>]

Hi Neal,


> On May 9, 2020, at 15:54, Neal Cardwell <ncardwell=[e-mail]> wrote:
> 
> On Sat, May 9, 2020 at 5:42 AM Jonathan Morton <[e-mail]> wrote:
>> [...]
> 
> It is not the case that L4S expects "the whole Internet" to be
> upgraded. AFAICT the L4S ecosystem is expecting something like the
> following:
> 
> *If* you are one of the few ISPs that has deployed RFC3168 ECN
> bottlenecks that do not have FQ or other mechanisms to ensure robust
> fairness properties in the face of flows that mark ECT(0|1) and do not
> follow RFC3168, *and* you are not comfortable with the L4S mechanisms
> to infer the presence of RFC3168 bottlenecks, then you should either
> (a) upgrade your RFC3168 ECN bottlenecks to something with more robust
> fairness properties (e.g. fq_codel or cake), or (b) simply change your
> existing DSCP rewriting rules at the ISP ingress to bleach ECT(1) to
> Not-ECT (or any other mechanism of your chosing to treat ECT(1) as
> Not-ECT).
> 
> neal

	[SM] IMHO this expectation really is demanding that others clean up the mess L4S will have made, no?
Other AQMs that presently work well and standards-compliant will need to change to allow L4S to work outside of existing RFC3168 rules. And that for rather little gains L4S offers over the state of the art ... (fq_codel based AQM can deliver a median of around 5ms under load with bog-standard 1/sqrt(p)type TCP flows, sure 1ms < 5 ms, and what ever many nines L4S secifies is larger than the median, but we are still talking relative small differences compared to the multiple 100s of miliseconds without any AQM, but that later is the wrong reference point to judge the L4S costs against IMHO)

	Your proposed remedy, using FQ to isolate the fall-out of L4S flows lack of CE responsiveness to themselves, is exactly what L4S itself eschews on the argument that FQ is computationally too expensive. Now, I am not convinced that that is actually technically correct (and assume L4S FQ-allergy is mostly driven by Bob Briscoe's known position on FQ) but still that is a self-inconsistent position to take IMHO: to allow L4S to get by cheaply, existing users are required to up their game at a considerable computational cost. 

I am not saying that this is not a valid position to take, but the consequences should be explicitly discussed.

Also, as a nit, the 2 ECN bits live outside the 6-bit DSCP field, so DSCP rewriting rules might not be the right term?

Best Regards
	Sebastian