Re: [tsvwg] L4S DSCP (was: L4S drafts: Next Steps)

[Sebastian Moeller <[e-mail]>]

Hi Ruediger,


> On Mar 24, 2021, at 17:05, <[e-mail]> <[e-mail]> wrote:
> 
> Hi Gorry,
> 
> One detail only, otherwise we are in line otherwise, I think
> 
>> Then innocent bystanders may get hit and a specified DSCP might allow for protection.
> 
> [GF] The bystanders are those impacted by a "CE-marking from a non-L4S router" that is congested, but where the other traffic responds using L4S.
> 
> [GF] Is this any different to the first operator not forwarding the ECT(1) codepoint?
> 
> [GF] I do wonder if this proves a real problem whether the community might deprecate CE-marking of ECT(0) in non-L4S devices, I wonder in future how much hardship this would cause, especially if L4S were to be widely deployed.
> 
> [RG] In the case of an operator sending L4S traffic, the source of a "CE-marking from a non-L4S router" must be upstream of the terminal. That is likely a policy point of the operator or an external network. The former doesn't apply and I doubt that the latter is to be expected. I'd expect the terminals to be deployed for some experiment (and users are either operator employees or friendly volunteers), so the contact with external networks sending "CE-marking from a non-L4S router" might be rare (I'd wonder about that - where's the bottleneck, a peering interface?).


	[SM] Almost, CableLabs has made L4S style signaling and AQM mandatory as part of Low Latency DOCSIS 3.1, the mandate seems to cover both the CPE as well as the CMTS. If the L4S experimental RFCs will be released it is very likely that DOCSIS ISPs will enable these features pretty quickly. After all, marketing for low latency docsis started years ago. IMHO this will en-roll quite a number of users into the experiment that are neither "operator employees or friendly volunteers". In that case A DSCP can do wonders to restrict the reach of L4S signaling to participating / L4S aware networks, as SLAs/TCAs are likely required to make sure that tose DSCP are passed on.

	As I have been told, the experiment in the 4S experimental drafts is not about safety or functionality, but about deployment, I expect mass deployment pretty soon. A DSCP during the experiment to reduce the "blast radius" seems like a cautious engineering approach to me, but then I am not an engineer. Also it will be considerably easier, in case I was too pessimistic, to remove the guarding DSCP requirement if L4S is successful and should be elevated to proposed standard status, than introducing one at that point, when mass deployment is already in process or even mostly finished.

Best Regards
	Sebastian


> 
> Regards,
> 
> Ruediger
> 
> 
> 
>