Re: [PANRG] About Keith and Michael's proposal on sidecar protocols in PANRG today

[Michael Welzl <[e-mail]>]

Apologies, I should have proof-read this email before sending:

> Aha. Well, to begin with, to a LIA or MPTCP sender-side CC instance,

Of course I meant “LIA or OLIA MPTCP sender-side CC instance”, and:


> things would seem normal: they would be delivered ACKs per path, and know what to do with them.
> If, then, the SC proxy would carry out its own congestion control downstream (like in use case #1 that I showed), then… yes there would be fairness problems, much like today’s MPTCP CC would get its fairness wrecked by a downstream connection-splitting PEP operating on only one of its subflows. If that’s a bit problem in practice is another question…

of course I meant a “big” problem.

Cheers,
Michael


>   I mean, if you try to make a transport work through a special link where e2e CC just doesn’t cut it (e.g., because the link’s capacity fluctuates), then you already face quite different problems than the long-term fairness between MPTCP sub-flows.
> Like, your cwnd has just been cut to almost nothing, and now capacity is available again: what’s the point of long-term (O)LIA fairness when, by the time your flow finishes, you just can’t get up to the available capacity anyway?
> 
> 
>> Traditionally of course a TCP PEP would just ignore this, since
>> it's kind of doing an undocumented hack anyway.
> 
> exactly -
> 
> 
>>  But with all the
>> explicit communication with the PEP in this scenario it might be
>> possible to do something, which might either make the docs (and
>> perhaps even the implementations and the signaling) somewhat more
>> complex or force a doc to acknowledge that the RFC series's views
>> on congestion control can be a bit of a naive fiction these days
>> in some regards (or both).
> 
> this sounds like we’re on the same page. I think being able to saturate capacities should be our primary goal, and maybe we should *really* care about long-term fairness in Congestion Avoidance once we have real-life sightings of this strange beast again (a CA phase that lasts more than a handful of round-trips, maybe even competing for capacity with anything).
> 
> But yes… we can surely add complexity to be theoretically sound  :-)
> 
> Cheers,
> Michael
> 
> _______________________________________________
> Panrg mailing list
> [e-mail]
> https://www.irtf.org/mailman/listinfo/panrg