[ih] "Gateway Issue": End-Middle Interactions

[Brian E Carpenter]

Karl,

There's inevitably a lot of end-to-middle in the ongoing work
on deterministic networking [1]. As far as I can tell from a
quick glance, they will likely use RSVP-TE [2] for signaling.

However, at Internet scale, I doubt we'll ever see any descendants
of RSVP. It would rely on stateful collaboration between ISPs
along the path, and that basically doesn't happen. The diffserv
model has some success across some ISP-ISP boundaries, but that's
stateless.

[1] https://datatracker.ietf.org/wg/detnet/documents/
[2] https://www.rfc-editor.org/rfc/rfc3209.html

Regards
    Brian Carpenter

On 07-Oct-24 09:04, Karl Auerbach via Internet-history wrote:
> On 10/6/24 11:55 AM, Jack Haverty via Internet-history wrote:
> 
>> ---------------------
>> Issue: End-Middle Interactions
>>
> In the mid/late 1990's Fred Baker and I worked on the router and client
> halves of the path resource reservation protocol, RSVP. That was a kind
> of end-to-middle interaction.  RSVP never caught on. But I do think that
> the idea ought to be resurrected in a modern form, but I have a nagging
> feel that even if we got the tech to work it would not be something that
> would be practical given the competitive nature of today's providers.
> 
> A bit later when I was trying to figure out how to do very fast and
> cheap discovery and binding of video clients to video services I came up
> with the beginnings of a protocol to evaluate hither-to-yon paths
> (including routing branches) in a bit more than one round trip time and
> with low priority processing in the routers along those paths.  I called
> it a "Fast Path Characterization Protocol" or FPCP.  (This was part of a
> contest I had with Bruce Mah - I wanted to see what information I could
> squeeze out of a position inside the routers and switches along the way
> while Bruce was seeing what he could get from the outside.  He was more
> successful than I was, partially because I tend to over-design and end
> up sinking myself into a swamp of code.)
> 
> The very sketchy draft of FPCP is at the URL below.  Although this work
> was done at Cisco I got permission to publish it.
> 
> https://www.cavebear.com/archive/fpcp/fpcp-sept-19-2000.html
> 
>           --karl--
>