[ih] QUIC story

[Toerless Eckert]

On Sun, Jun 26, 2022 at 06:04:48PM +0200, Stephane Bortzmeyer wrote:
> > the long-perceived wisdom was that security, such as via TLS or
> > IPsec is best decoupled from transport to be reuseable/modular. And
> > that QUIC experience may serve as a reminder that other real-world
> > requirements may throw wrenches into other well-meaning modularily
> > and layering principles.
> 
> Another way, more positive, to view this change is to say that
> encryption, today, is no longer an option, it is as necessary as flow
> control and congestion avoidance and, therefore, it makes sense to
> have it inside the transport layer.

Even if encryption was optional, you wouldn't want the RTT overhead
that decoupled transport and security layer would give, when you do
use encryption, so ultimately it is the need for speed (low RTT) that
drives breaking traditional layering assumptions. Encryption is just
one example where this is true. And of course for encryption it doesn't
only happen with transport.


Low overhead for high performance at low cost are just two IMHO in our (protocol)
world architecturally underappreciated ongoing challenges that in practice
have been at the forefront of driving adoption and proliferation of
our protocol technologies. I would not be surprised to see also an
ongoing trend to see our existing IETF protocols be superceeded by
the more lightweight variants we have been building especially over
the last decade (as long as those alternatives have no significant
use-case limitations).

That bad part of this is that its extremely difficult then to
future-proof protocols through expandability, because that runs
quite contrary. Theres some good IAB insight written recently on that
too.

> The principle of layering is very important. But the actual placement
> of layers can vary.

-- 
---
tte at cs.fau.de