Re: Limited Domains:

[Tom Herbert <[e-mail]>]

On Wed, Apr 28, 2021 at 4:59 PM Toerless Eckert <[e-mail]> wrote:
>
> On Wed, Apr 28, 2021 at 10:51:03PM +0200, Tony Przygienda wrote:
> > yepp, researchers are having fun unencumbered largely by harsh realities
>
> Well, the better researcher attempt to get more of an understanding
> of the harsh realities by working with whats available to them, which is
> quite limited - a few P4 platforms. This lack of openness to multi-party
> innovation is what causes high-speed router platforms to become less and
> less relevant and devolve into commodities, at least at the level where
> folks in IETF are interested in them. Obviously the spped&feed hardware
> level will stay an ongoing motor of innovation and maybe even sales
> margin, but for the packet layer forwarding, we can close shop if we do
> not have a strategy to allow new markets/innovation to be brought in.
>
> Heck, take a look at the challengers in 5G/6G. They are already moving
> purely into general purpose CPU because those challengers have no way to
> put innovation into accelerated forwarding hardware due to lack of access,
> complexity and also being stuck in a world of standards such as GTP that make
> innovation difficult.
>
Toerless,

While I agree with your sentiment, this is really a lament about a
segment of the market not about anything IETF has done. For instance,
no IETF standard  requires high performance routers to be implemented
in fixed ASIC, in fact I believe the technology has evolved to the
extent the datapath could go through the CPU simultaneously achieving
both performance and flexibility (my company is working precisely on
that).

The hard problem is legacy and the Least Common Denominator effect. On
the Internet we are bound by the least common denominator of protocols
which is plain TCP/IPv4 and TCP/IPv6 that generally make it through
the Internet. UDP is looking a little better, at least because of QUIC
which took a behemoth to force support. The end result of all this is
the ossification of the Internet which as you point out stifles
innovation and creation of new markets. This is actually the appeal of
limited domains, it creates walled gardens where in which the least
common denominator can be raised and we can innovate in that sand box,
however limited domains don't address the underlying issue unless
their scope continually expands to cover more and more of the
Internet.

Tom

> Its totally incomprehensible to me, why engineers who paycheck ultimately
> depends on hardware that already is way more flexible than what our current
> protocols use do not try harder to think about better models for third parties
> to get more value from this hardware. More flexible, programmable packet forwarding
> options are one core option here. And obviously no sane customer
> wants something thats not multi-vendor in this space which is where
> IETF could come in. And no sane network operator wants another protocol silo.
> Which is where IPv4/IPv6 compatibility comes in as a base level expectation.
>
> > while meanwhile large scale IP forwarding is still governed by the
> > microcents/bit payload that customers can run their business profitably at
> > and this is governed in turn by the merciless trifecta of physical laws of
> > power/cooling/space that did not only not change in the last 50 years but
> > last 500 and probably much longer even before Newton laid his cranium under
> > the said apple tree ...
>
> When single-family home systems are built from 19" CPU server racks
> providing compute across FTTH and powered by solar electricity, i think
> we are also starting to see big shifts in the economics of those core
> factors. But thats digression.
>
> I think we agree that from the protocols we remember, IP/IPv6 has
> done a pretty good job providing value out of the hardware
> deployed. But we also know that other protocols like Ethernet and MPLS
> have shown other sectors where they are doing better. So it seem to me
> obvious that as developers we should look ahead and see what it is we
> can best do to crearte more value. And certainly we know our hardware
> can already do more than what is utilized by the protocols.
>
> Cheers
>     Toerless
>
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [e-mail]
> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------