[ih] Internet without entrenched factions?

[Brian E Carpenter]

I find myself scratching at this itch. Having failed to learn from history, are we repeating it? It turns out we have data on that (thanks to Geoff Huston). Look in particular at the third graph at https://cs.auckland.ac.nz/~brian/sqlaw.html. Although it ends with a 2025 data point, I created it a few minutes ago.

The bottom line: IPv6 prefixes don't seem to be aggregating any better than IPv4 prefixes did ~ten years earlier.

Regards/Ngā mihi
    Brian Carpenter

On 19-May-26 16:22, Brian E Carpenter wrote:
> Thank you, Tony.
> 
>> 1M IPv4 prefixes and 0.25M IPv6 prefixes.
> 
> In fact, since it appears that at least 50% of end hosts now have IPv6
> connectivity, that isn't as bad as it might have been. But of course few
> of them are multihoming.
> 
> However, the last time I looked (in 2020), things were definitely
> getting relatively worse.
> (Fig. 3 of https://cs.auckland.ac.nz/~brian/sqlaw-revisited-published).
> 
> Regards/Ngā mihi
>      Brian Carpenter
> 
> On 19-May-26 16:02, Tony Li wrote:
>>
>> Hi all,
>>
>>>> PS: However, you didn't explain what the speed of light has to do with
>>>> why it's too hard to fix multihoming... nor IPv4.
>>>
>>> It's to do with the fact that if the rate of BGP4 updates increases
>>> dramatically, as it would if we were trying to multihome tens of
>>> millions of enterprises, they would still have to be propagated
>>> at the speed of light. Moore's law could presumably provide for
>>> storage and processing of those millions of BGP routes but only
>>> after the updates make it to every default router. It's unclear
>>> that BGP routing could ever converge at that scale. Tony Li
>>> should be answering your question, not me, however.
>>
>> Sorry for the delay, I’ve been travelling.
>>
>> Multi-homing is not hard to fix.  You just have to do it properly.
>>
>> If you want multi-homing to scale, you have to avoid having every multi-homed site resort to PI space and their own prefix in the DFZ.  BGP converges sub-linearly with the number of prefixes in the DFZ and overloading BGP is not going to help anyone or anything. If you care about the details of path hunting and flap damping, we can get into that.
>>
>> Thus, if a multi-homed site is not going to use PI space, then it needs to be part of multiple provider prefixes. ILNP showed us a clean way of doing that, but it was too much of an architectural change for us to swallow.
>>
>> So here we sit. People abuse the DFZ with PI space, and are now talking about attaching PI prefixes to each and every edge router. Pollution due to deaggregation for inbound traffic engineering purposes.  1M IPv4 prefixes and 0.25M IPv6 prefixes.  And no way of saying ’stop’.
>>
>> It has been my privilege and honor to serve for lo these many decades, but when these chickens come home to roost, I will be retiring.
>>
>> Regards,
>> Tony
>>