Re: [arch-d] FYI: closure of the IAB Stack Evolution program

[Fernando Gont <[e-mail]>]

On 26/8/19 08:24, Joe Touch wrote:
> The only thing QUIC doesn’t do is demux via a single field of the IP
> header called “protocol”.

I go back to my original comment: that you can't deploy new transport
protocols. Yes, you can stack as many layers as you want, but you can't
deploy new transport protocols. You can deploy new transport protocols
as much as you can deploy new internet protocols (you can do overlays as
much as you want). To put it bluntly: you have UDP sitting below QUIC
because you can't deploy new transport protocols on the Internet.

(No need to mention that none of my previous comments was in detriment
of QUIC -- they were very pragmatic, indeed).



> Maybe this is a good thing - maybe it just goes to show middleboxes how
> incapable they are at actually knowing what’s going on.

I'd bet much of the brokenness has nothing to do with middle-boxes that
want to know what's going on (e.g., firewalls enforcing security
policies). A lot has to do with NATs (which need to be aware about the
transport protocol to do the translation), and other boxes that at times
need to drop packets as a side effect of doing something else
(https://tools.ietf.org/html/draft-gont-v6ops-ipv6-ehs-packet-drops).

Let's also not just blame NATs: we were extremely close to ship what
became RFC8200 with no comments about middle-box insertion of EHs, we
also had SRv6 with in-the-network EH-insertion, and Linux currently
ships with in-the-network SRv6 insertion (see man 8 ip-route). At times
we don't seem to be consistent regarding wether we want middleboxes to
fiddle with packets or not....

-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492