Re: [tsvwg] L4S vs SCE

[Jonathan Morton <[e-mail]>]

>>Where I think SCE fails is that it is not available to any link that
doesn't implement FQ, whether that is by choice or by necessity.
>
>         [SM] SCE is a marking strategy pure and simple. I see no
insurmountable obstacle in combining the backward compatible SCE marking
strategy with a better classifier to give L4S queueing to selected traffic.

I prefer to describe SCE as a signalling mechanism, one which can support a
variety of marking strategies (e.g. threshold, RED, Codel) and response
functions.

That is the fundamental difference here, and I think that L4S could
reasonably switch to SCE signalling with essentially all of its advertised
goals intact.  That would be a robust way to close several of the
outstanding issues in the tracker.

I should also note that whenever there is unfairness between SCE and
non-SCE flows, it it's in favour of the non-SCE flows.  This means SCE is
safe to deploy even in a single queue; it just might not perform well.  The
reverse is true of L4S, which is why it needs protection for safe
deployment unless and until the RFC-3168 detection heuristic is shown to
work robustly.

- Jonathan Morton