Re: [arch-d] Fwd: A Public Option for the Core

[Toerless Eckert <[e-mail]>]

On Mon, Aug 17, 2020 at 01:55:13PM +0000, Andrew Campling wrote:
> 
> > On Aug 17, 2020, at 03:46, Toerless Eckert <[e-mail]> wrote:
> > 
> > John, Joe, Brian:
> > 
> > All great discussion points, but i am not sure if JohnD's starting 
> > point of "no deductive adaptation of behavior" is really mandatory 
> > and/or sufficient for *I* net neutrality. Aka: Do we even have an 
> > approved and complete *I* "net neutrality" definition ? If so, where ?
> 
> You seem to be assuming that "net neutrality" is always a good thing?  My personal preference is for networks to be transparent in the way that they manage traffic (if at all) so that I can choose my network based on what they do.  A network without, for example, any form of QoS might perform very poorly for at least some applications.  

I was only asking for definitions of net neutrality.

I fully agree with your statement. I hope there is no definition of
net neutrality that would prohibit such differentiation of service
level objectives of different traffic (which hopefully is a broder
and more accurate term than QoS).

> > Btw, IMHO: Having an application tell the network exactly what it 
> > wants from the network is quite a challenging task for applications.
> > When i worked with enterprise application developers, they wouldn't 
> > even want to bother to understand what differences in service from the 
> > network would best help the application.
> 
> This point I agree with.  In general, the level of knowledge of networks by applications developers can be very low (and vice versa), they exist in very different worlds.  

Most advanced SLO (service level objective) knowledge IMHO can be found
in applications resulting from vertically integrated development. Aka:
Applications written on behalf of an organization that is capable and
happy to provision network services to support SLO that makes the
application perform best: enterprises, self-owned data-centers, defense,
space, IoT (transportation, oil&gas, building integration,...).

service provider networks only came into play when all those mission
centric networks where outsourced or had to be merged into common
infrastructures. Forgetting this history is one of the key reason
why we are struggling of driving QoS/SLO support forward.

> > In classical enterprise environments, this trusted mapping layer could 
> > typically be a nework device (given how applications in enterprise are 
> > subject to a lot of control by the enterprise operator anyhow).
> > In typical residential environment, the trusted mapping would likely 
> > best be the browser, given how that performs all type of control 
> > operations against applications anyhow.
> 
> Whilst I believe that I understand your point, I'm not at all convinced that, in residential environments, delegation of trusted mapping to browsers makes a lot of sense.  That seems to present a fine attack service for future exploitation.  No doubt others will have different views though.  

Browsers do provide  a  well constrained  environment for applications,
and there is  an active SDO (w3c) defining those standards. Even with
some help from IETF (RTCweb for example). Thats why purely from the
perspective of an SDO, the chance to define standards is better for this
option.

If you take a random app
and tried for the OS to do the mapping, there is no standard.For example,
microsoft had a great mapping scheme as early as windows XP (or before,
i only started to use it with XP) to trigger diffserv/intserv. configured
for different apps by the windows administrator (e.g.: across an enterprise). 

I have seen enterprises well utilizing this windows infrastructure.
Alas, AFAIK, this never trickled through across other OSs because 
obviously Microsoft tried to leverage this as a competitive differentiator,
and the more application traffic became Internet centric, the less
relevant this became.

Likewise, all DPI/QoS mapping in routers is of course also vendor
proprietary, even though there is of course a lot of commonality across many
vendors.

Aka: From the perspective of an SDO the question is firt what the
architectural APIs are to be defined. The instances through which the
mapping could be done (browser,OS, router, other?) could then even be
proprietary adaptations (ideally standardized too).

Cheers 
    Toerless

> Andrew
> 
> -----Original Message-----
> From: Toerless Eckert <[e-mail]> 
> Sent: 17 August 2020 08:47
> To: Brian E Carpenter <[e-mail]>
> Cc: Joe Touch <[e-mail]>; John Day <[e-mail]>; [e-mail]
> Subject: Re: [arch-d] Fwd: A Public Option for the Core
> 
> John, Joe, Brian:
> 
> All great discussion points, but i am not sure if JohnD's starting point of "no deductive adaptation of behavior" is really mandatory and/or sufficient for *I* net neutrality. Aka: Do we even have an approved and complete *I* "net neutrality" definition ? If so, where ?
> 
> Btw, IMHO: Having an application tell the network exactly what it wants from the network is quite a challenging task for applications.
> When i worked with enterprise application developers, they wouldn't even want to bother to understand what differences in service from the network would best help the application.
> 
> Hence we thought of a framework where the application would instead attribute traffic as to what it is, e.g.: application FOO, wihthin FOO floX,
> e.g.: X = audio/video/...  Aka: Whatever the application developer was able/ willing to tell.
> 
> Then one can have a trusted layer that would map this information to actual network service requirements/advisory information.
> 
> In classical enterprise environments, this trusted mapping layer could typically be a nework device (given how applications in enterprise are subject to a lot of control by the enterprise operator anyhow).
> In typical residential environment, the trusted mapping would likely best be the browser, given how that performs all type of control operations against applications anyhow.
> 
> Cheers
>     Toerless
> 
> On Mon, Aug 17, 2020 at 08:24:06AM +1200, Brian E Carpenter wrote:
> > On 17-Aug-20 03:01, Joe Touch wrote:
> > > And the second is that service level implies connection. It only requires the network understands a label *I* control, rather than trying to infer one from the packet contents.
> > 
> > However, it does need some kind of flow state (whether flows are individual or aggregated) in the forwarding queues. Not that this is a new argument; we had it in the diffserv WG many years ago and it continues to this day in TSVWG.
> > 
> > I don't think this really affects the basic arguments for a "public core". It just indicates that transit networks don't need to be strictly best effort and stateless in order to be neutral.
> > 
> >    Brian
> >  
> > > 
> > > Joe
> > > 
> > >> On Aug 16, 2020, at 3:19 AM, John Day <[e-mail]> wrote:
> > >>
> > >> ???
> > >>
> > >>> Begin forwarded message:
> > >>>
> > >>> *From: *John Day <[e-mail] 
> > >>> <mailto:[e-mail]>>
> > >>> *Subject: **Re: [arch-d] A Public Option for the Core*
> > >>> *Date: *August 16, 2020 at 06:18:04 EDT
> > >>> *To: *John Grant <[e-mail] <mailto:[e-mail]>>
> > >>>
> > >>> No, that is not true.  Your first mistake is that there is a control plane.
> > >>>
> > >>> John
> > >>>
> > >>>> On Aug 16, 2020, at 06:02, John Grant <[e-mail] <mailto:[e-mail]>> wrote:
> > >>>>
> > >>>> On 16/08/2020 02:48, Joseph Touch wrote:
> > >>>>> *I* want apps to be able to get different service from the network and to pay differently for them if needed, but *never* to have the network infer or enforce that mapping.
> > >>>> Of course, if you're going to do that -- have the application negotiate a particular kind of service rather than have the network "inspect" the packets and guess what service is needed -- you need a control plane protocol (let's call it "signalling") to do the negotiation, and you no longer have the connectionless/stateless paradigm on which IP is founded.
> > >>>>
> > >>>> --
> > >>>> John Grant
> > >>>> Nine Tiles, Cambridge, England
> > >>>> [phone] and [phone]
> > >>>> http://www.ninetiles.com
> > >>>>
> > >>>> _______________________________________________
> > >>>> Architecture-discuss mailing list [e-mail] 
> > >>>> https://www.ietf.org/mailman/listinfo/architecture-discuss
> > >>>
> > >>
> > >> _______________________________________________
> > >> Architecture-discuss mailing list
> > >> [e-mail]
> > >> https://www.ietf.org/mailman/listinfo/architecture-discuss
> > > 
> > > _______________________________________________
> > > Architecture-discuss mailing list
> > > [e-mail]
> > > https://www.ietf.org/mailman/listinfo/architecture-discuss
> > > 
> > 
> > _______________________________________________
> > Architecture-discuss mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/architecture-discuss
> 
> --
> ---
> [e-mail]
> 

-- 
---
[e-mail]