[tcpPrague] Too fast for Google?

[Bob Briscoe <[e-mail]>]

Jana,

When we held the L4S Bar BoF in Buenos Aires, you said we were moving 
too fast.
I think the sentiment behind your point was that we seemed to be 
shutting out similar approaches too fast.
Can you clarify if that's what you meant?

If yes, I'd be interested to know if this impression came from the 
presentations or the drafts. Because:
* in the drafts, we've specified example design(s) in appendices, and in 
the normative body text, we've distilled only the essential structure of 
the approach as normative standards statements.
* in presentations, I (at least) have a tendency to want to be concrete 
- to demonstrate that there is real code and equipment and tests, etc.

I'd also be interested to hear if you think we have been successful in 
generalising the drafts, or whether you think they are still too specific.

In particular, you mentioned making space for delay-based congestion 
control, which I would /not/ want to include for the reasons below.
Is there any rationale for being /that/ general?

Background:
L4S aims for a very shallow queue, which we indicate with immediate ECN 
notification (the L4S demo used a 5-MTU threshold when there is no 
traffic in the classic queue).
ECN gives much more precision, and immediately. Rather than waiting for 
enough packets to get an imprecise delay-measurement heuristic.
(for instance the sender can dither packet inter-arrival time during 
flow-start to sense available capacity, which it can then approach very 
rapidly without overshoot.)

This is what I meant when I said we have created an "incrementally 
deployable clean-slate"...
There is a brief window of opportunity during which we (ie. you, 
everyone) can all jiggle both host and AQM together. But it will have to 
be standardised soon, then it will ossify. We can't wait too long, 
otherwise everyone loses interest.


Seriously, I first presented the idea of deploying DCTCP on the Internet 
in the same session as Van Jacobson introduced CoDel (Jul 2012). We 
published the main drafts on L4S in Jul 2015, so I think it's reasonable 
to expect to be moving on to setting standards agendas about a year 
after that.
Your thoughts are always enlightening, so if you haven't set aside time 
to read, consider deeply and comment,... soon would be good.



Bob

[I actually wrote a similar email straight after Buenos Aires, but I 
just noticed I sent it from the wrong address, so the list rejected it.]

On 01/06/16 11:53, Bob Briscoe wrote:
> All,
>
> Can some folks respond on whether they think it is right to hold a BoF 
> on this topic in Berlin.
>
> Back in Jul 2015, when we had the first ad hoc meeting about evolving 
> DCTCP (the one where Matt Mathis suggested we call it TCP Prague), 
> there was a huge amount of energy and excitement in the room.
>
> However, I've noticed that the discussion on this list seems to move 
> in fits and starts.
> Can anyone suggest why?
>
> I see much more discussion off-list about building projects, around 
> the enabling idea of the dualQ.
> Perhaps it's just that it takes a long time for some people to be able 
> to to shift the focus of their work.
>
> If that's so, then I guess it is still worth the core proponents 
> pressing ahead.
> Because others only need to influence the direction of travel of a 
> project if it is moving!
>
> Thoughts?
>
>
> Bob
>
> PS. The BoF proposal deadline is on Friday.
> I'm guilty for not having yet delivered what I promised: a problem 
> statement draft - working on it now.
>
>
>
> -- 
> ________________________________________________________________
> Bob Briscoe                               http://bobbriscoe.net/