Re: [tsvwg] Dual-Q Polcing

[Gorry Fairhurst <[e-mail]>]

On 25/05/2020 11:10, Jonathan Morton wrote:
>> On 25 May, 2020, at 1:00 pm, Sebastian Moeller <[e-mail]> wrote:
>>
>> Is everybody else convinced that L4S offers a sufficiently safe design and was tested with the to be expected level of adversarial traffic patterns to confirm the design's safety in the real-world?
> I am convinced of the opposite of this.  Hence the many technical concerns raised over the past few weeks.
>
> Some results that Pete Heist posted recently demonstrate, if proof were needed, that traffic unresponsive to ECN signals can easily crowd out responsive traffic flows in DualQ, whereas straightforward implementations of flow-aware queue protection (ie. the BLUE component of COBALT used within CodelAF) are able to mitigate that sort of behaviour, even without an explicit FQ mechanism.  Earlier results also show easily triggerable bad results without requiring any intentional misbehaviour.
>
> That, I think, lies at the root of Sebastian's line of argument.
>
>   - Jonathan Morton

Jonathan,

To me these seem like examples of reasonable operational tools when a network wants to provide flow-level management. Somewhere the documentation produced by the IETF should indeed describe this type of mechanism and how they can be deployed to manage traffic unresponsive to ECN signals. And I'd expect there are likely also other ways also to mitigate the risks from unresponsive traffic, and it could be posisble to offer a range of methods and decsribe their merits?

Gorry