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

[Fernando Gont <[e-mail]>]

On 25/8/19 21:07, Christian Huitema wrote:
> On 8/25/2019 6:59 AM, Fernando Gont wrote:
> 
>> I agree with the above, still, the reason for which QUIC employs UDP
>> (rather than doing mux/demux by itself) is because you can't deploy new
>> transport protocols on the Internet.
> 
> 
> I think we are getting caught in disputes over vocabulary. Two things
> are true at the same time:
> 
> 1) We cannot send packets in which the IP protocol or IPv6 payload type
> is set to a different value than TCP or UDP and expect them to reach
> destinations across the Internet.

Agreed. But doesn't that mean that you cannot deploy new transport
protocols natively?

Architecturally-wise, in a non-recursive architecture transport runs
over Internet, right?

If IP is Internet, and QUIC is transport, what do you call the UDP that
stis in between?



> 2) QUIC, BitTorrent and SCTP demonstrate that you can run a protocol on
> top of UDP that has all the characteristics of a classic transport
> protocol, including error recovery and congestion control. ( QUIC can
> perform mux/demux by itself, using the "connection ID" field, but most
> deployments refrain from using multiple layers of multiplexing.)

As long as UDP is not filtered, you can run anything you want on top of
UDP. You can run IP or IPv6, too (I regularly do, indeed). (sixxs has
done that for ages, openvpn and IPsec over UDP, too, and I seem to
recall that sctp too)



> Using UDP causes 8 bytes of overhead per packet, which is unfortunate.
> You may call that a middle-box tax if you want. But as long as you are
> willing to pay that tax, you can of course deploy new transmission
> control protocols over the Internet. Most people will call them
> transport protocols.

The point of my original comment was this:

1) You cannot deploy new transport protocols natively. You need to
encapsulate/tunnel, because otherwise the big-i internet will likely
drop your packets. This is kind of well understood.

2) Of course you can always find some protocol that will work, to tunnel
something you want to deploy (go up one layer if the current layer
doesn't work). So I'm not sure there's any news here. -- that doesn't
mean less credit for QUIC at all. I'm wondering if the fact of being
able to find *something* over which a protocol (transport or otherwise)
can work has been a new key finding. -- e.g. any of us have been doing
OpenVPN over UDP and others for quite a while.

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