[ih] Better-than-Best Effort

[Toerless Eckert]

On Thu, Aug 26, 2021 at 03:39:54PM -0700, Dave Crocker via Internet-history wrote:
> Having not followed actual QOS work over the year, my naive brain wandered
> oddly, today with a thought about a semi-QOS approach.

I think the unit of observable changes in QoS is typically greater than a decade.

Then again, L4S is making slow? progress, NADA became RFC in 2020 and DetNet
released its first significant round of RFC end of 2019, except that it has
no good wide-area nework queuin solutions, hence i wrote draft-eckert-detnet-bounded-latency-problems
and i am sure i am not tracking jus a small part of what may be going on.

> The usual view is that it requires complete, end-to-end support. Massive
> barriers to adoption, at the least.

Depends probably on how-much-better-than-BE it intends to be.

> I'm thinking that the long-haul infrastructure tends to have enough capacity
> that it usually isn't the source of latency.  It's the beginning and ending
> legs that do.

Not necessarily:
There should be a good amount of OTT nowadays that will show you that they can
sell you paths across the planet with better (stochastical) latency bounds than
native Internet paths. Most of them do not even own any links, but just tunnel
across various hops with container/VMs that they pay for.

But i do agree that the mayority of future use-cases that go beyond the current
application success story of the Internet (non-real-time, non-critical) will 
likely fail on the unreliability and imprecise latency bounds specifically in what
 i would call the metropolitan edge of the Internet. Aka: new applications running between
edge-data-centers in a metropolitan region and subscribers in the same region.

You can read my architectural view of that startin on page 103 of:
https://www.itu.int/en/ITU-T/focusgroups/net2030/Documents/Network_2030_Architecture-framework.pdf

> So what about a scheme that defines and provides QOS in those segments but
> not the long middle?  Cheaper, more implementable, and might give
> usefully-better performance.

PIE only made it through IETF because DOCSIS wanted it, so we do see it
in that metro edge. The current/next generation of even whitebox metro switches
do seem to start getting AQM of similar strength. Aka: The scheme that works
is ad-hoc introducing of AQM into your very own congestion points to improve
your subscribers experience. I always say BE before AQM was LE (Lousy Effort).
So at least with AQM and better CC we are closer to a least getting BE ;-)

I think where we hit the brick wall to adopt any more of any better _differentiated) QoS options,
maybe starting with L4S, but hopefully also metro-size deterministic netwrking for
remote-driving and live&death reliable services, is in any form of business
models that would ask and pay for such network functionality.

Pessimistic as i am,
I think those business models will again, like we saw 20 years ago with MPLS/VPN
evolve in isolated VPN/slices across the same infrastructure. And because they
are driven by a small number of customers such as mobile operators, industrial or public
services/traffic-control/power-distribution/... etc, we will just see a proliferation of hacked-together
qos for one-off solutions. Like i have seen it in QoS in MPLS/VPN. Managemenet
of Queue weights by FAX messages between customer and subscriber is my favourite common hack.

As an ex-colleague-liked to say: www.showmethemoneyforqos.com

> Assuming that this idea is new only to me, I'm curious about
> reactions/history/etc.

AFAK, IETF has not done any real QoS architecture since the 90th, except for
now DetNet, which is IMHO very badly attended by the mayority of vendors
or operators, and can IMHO also onlyy be the most high-end niche solution.
Instead, everybody is optimizing in their own bubble where they can. Mostly
with positioning of compute, a bit of AQM, better CC and increasing capacity
through peering.

Remember also that better QoS support in the Internet for applications immediately
also get the whole brigade of general concerns about privacy and net neutrality
raised against it by those that likely make the most money of better traffic
experiences in their own, non-public-internet service offerings. So i fear better
QoS architectures for the Internet are also a victim of commercial interests that
ultimately will make the Internet and its high level of oversight get replaced
by various propriety network service offerings with better QoS.

Cheers
    Toerless

> d/
> -- 
> Dave Crocker
> Brandenburg InternetWorking
> bbiw.net
> -- 
> Internet-history mailing list
> Internet-history at elists.isoc.org
> https://elists.isoc.org/mailman/listinfo/internet-history