Hi Michael,
Thanks for kicking this off. It's an interesting way to look at Geoff's presentation. Please see inline.
Best,
Dirk
-----Original Message-----
From: Semantic Address Routing and Hardware - SARAH [mailto:[log in to unmask]] On Behalf Of Michael Richardson
Sent: 22 November 2021 15:46
To: [log in to unmask]
Subject: [SARAH] the end of the Internet according to Geoff
(Interacting with the LISTSERV@JISCMAIL has taken me back to 1986, but somehow without the UUCP bangpaths...)
I have been thinking a lot about Geoff's presentation
_Centralization of service provisioning and the impact on routing_ (I wonder if he'll be repeating it at RIPE83 this week) I also happened to pick up Carl Malamud's 1991 _Exploring the Internet_ based upon some comments in [log in to unmask] I never saw the book at the time it came out.
It includes lots of interesting details and interviews with people.
I never knew Christian Huitema was an X.400 expert, for instance.
But, in particular, I didn't know, from my vantage point at lsuc!ncraer!fts1!latour!mcr that UUCP was so well used over X.25 networks at the time.
There is a connection here to Geoff's end of the end-to-end Internet, but I haven't quite figured out what it is yet.
Geoff's observations are that the CDNs have risen and that we really don't need end to end connectivity: just connectivity to the nearby CDNs.
[DOT] I contest this. If we believe various data sources, it is a LOT of traffic but I would argue that even just 10% of non-CDN traffic is still a lot (numbers seem closer to up to 70% being CDN served, leaving 30% on the table for E2E). Trouble here is: can we easily judge which traffic of mine belongs to which bucket? The question was asked in the chat (don't recall exactly who did, it may have been Tony) of the side meeting, that if transit was dead, why not switching it off and wait until the world screams?! While I am critical of the centralization, I don't see how this is the ruling architecture for overall Internet communication.
We don't need globally unique addresses, just locators that are unique to our local CDN. One imagines some other Internet evolution which assumed CDNs, and instead of doing CIDR and carving up the class As, if we had instead declared 0.0.0.0/1 as being RFC1918-like, and repeated it within each CDN "district".
Class B space would have been for (some) Enterprises and Governments to use to talk to each other, and I think class C space would be used by the CDNs to number themselves, essentially in an anycast like way.
Would IPv6 even be needed?
[DOT] No, not if connection to a central point is all we need - but see above on that assumption.
I have been thinking about what's stopping us from going there now.
I wonder, if one removes video stream, of the remaining traffic, what has still been growing, and what is dying? Imagine if we didn't have eyeballs capable of watching video, would the other services of the Internet pay for the numbering, naming, etc. infrastructure?
What are the things that make end-to-end worthwhile keeping?
[DOT] the non-POP traffic, which still seems to amount to quite some amount? Beyond that, I think it is also worthwhile asking what use cases may simply not well served by the central POP model. It is clearly the model to which a number of current use cases may converge to, and that seems sensible, but caution may be needed to assume that use cases beyond those we have now will be best served by it.
So my proposal is to have some kind of workshop, beer-bof in biergarten, or maybe online, where we attempt to design and debug the GH-CDN internet.
(Maybe Geoff would like to give this new architecture a name that does not involve his name...)
[DOT] I like this suggestion for the purpose to outline what we have right now and what we may want going beyond today. It aligns with a comment Dino made in the Int area side meeting that we may want to have a discussion on the features that we would expect the network to provide; not all may warrant a centralized type of realization that Geoff has observed in today's Internet.
Let's conjecture that the model that he presented is a good model and something that we *want*. Suspend your distaste and think like the "enemy"
First, what should it look like?
Second, how do we get there? What are the hurdles to getting to an all CDN situation? Who wins? Who loses? What savings do we see?
What new costs are there?
[DOT] To put it differently, I would propose an analysis of the gaps that this model was expose.
I think that there will be many surprises.
End-to-end things we didn't know were critical, and that we'd have to figure out a way to do differently.
[DOT] I tend to agree on that!
--
] Never tell me the odds! | ipv6 mesh networks [
] Michael Richardson, Sandelman Software Works | IoT architect [
] [log in to unmask] http://www.sandelman.ca/ | ruby on rails [
########################################################################
To unsubscribe from the SARAH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=SARAH&A=1
This message was issued to members of www.jiscmail.ac.uk/SARAH, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
########################################################################
To unsubscribe from the SARAH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=SARAH&A=1
This message was issued to members of www.jiscmail.ac.uk/SARAH, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
|
