Re: QUIC ossification

[Ted Hardie <[e-mail]>]

On Wed, Feb 13, 2019 at 3:47 PM Martin Thomson <[e-mail]> wrote:

> On Thu, Feb 14, 2019, at 04:17, Ted Hardie wrote:
> > On Wed, Feb 13, 2019 at 9:14 AM Mikkel Fahnøe Jørgensen <
> [e-mail]>
> > wrote:
> >
> > > There is the risk that middle boxes learn that versions mean nothing,
> and
> > > that they are all the same, akin to crying wolf.
> >
> > I tend to agree with this. Version numbers should shift when the
> > interoperability among parties to the protocol shifts.  While you can
> > artificially create that shift in closed system, teaching the ecosystem
> > that the shifts are meaningless doesn't seem to advance the general goal.
>
> If I understand David's proposal, these versions would have meaningful
> changes, or at least changes that go beyond the version number.  In TLS he
> has suggested a similar technique, with changes that are not compatible,
> but relatively easy to implement.  For instance, changing the salt for the
> Initial keys would be trivial to implement and make the protocol
> indecipherable.


Changing the salt is the kind of thing I had in mind for "artificially
create that shift."  That's easy to do in a closed system; it's possible in
a standards-based one, but the overall impact on functionality is so low,
that the rate of uptake will be low.  So you will do version negotiation
down to the old version a lot of the time, I'm guessing.


>   Switching to ChaCha20-Poly1305 instead of AES-128-GCM would be another.
> He also suggested that we reorder key fields, switch endianness of various
> encodings, rot13 extension codepoints, ... plus whatever you can imagine.
> As long as it is easier to respond correctly to new versions than it is to
> build specific handling for every new permutation, the protocol designers
> win.
>
> Now, I'm not sure if that is as good as having a good reason to deploy a
> new version, but there you are.
>
>
As a personal opinion, I think we have enough semantic additions to do here
that we don't need an artificial version which will have no natural driver
of adoption.

Ted