[IPv6]Re: Mail regarding draft-mishra-6man-variable-iids

[joel jaeggli <[e-mail]>]

On 9/17/24 04:34, Simon wrote:
> On this question I’m in danger of splinters from sitting on the fence. But there’s one statement I do strongly disagree with.
>
>
> On 12 Sep 2024, at 19:16, Mark Smith <[e-mail]> wrote:
>
>> People prefer single size edge subnets, as shown by how common IPv4 /24s are.
> I do not think the prevalence of /24s means people prefer single size subnets - I think it’s more a case of a) the default, and b) not knowing any different.

There are a lot of structure reasons why /24 subnets are normative 
structurally that go all the way back  to class c netblocks, a byte 
being 8 bits and so on. there's a minimum atomic size of a subnet that 
has a varaible number of devices on it that has existed as long as we've 
had bridging and while  it's possible that a /24 is to small for your 
needs there's a lot fewer cases with private IPs to think about is a /27 
or a /26 going to accommodate all the devices that may go on this 
subnet. these are way different considersation than the one were you 
might want a /64 or /120 or a /127  or something else.


>
> If we take a look back in time, RFC1918 https://datatracker.ietf.org/doc/html/rfc1918 defined the 192.168 block of private addresses in terms of "and third block is a set of 256 contiguous class C network numbers”. So back in the days when glassful addressing had not gone away, by definition, 192.168 subnets would be /24 in CIDR terms.
> So the default was to use /24, and routers implemented it by default - in fact I’ve worked with consumer/small business routers that could not cope with anything else, to the point of it’s GUI declaring x.x.x.0 or x.x.x.255 as not being valid IPv4 addresses !
>
> Since, by observation, the majority of people working with networks have learned at the “monkey see, monkey do” school, they then accept /24 as being the default. To the extent that I once worked with some colleagues who were adamant that I was breaking their imposed numbering plan by not reserving 172.16.1.1 through 172.16.1.10 in a /23 (in my bit of the group, we’d outgrown a /24) for network devices such as routers.
>
>  From the same “monkey see, monkey do” school, I’ve met people who declare that it’s not valid to host any device on a public IP - it must be behind NAT to be valid. And don’t get me started on how DHCP delivered addresses are supposedly “fixed” in Windows networks - until they aren’t and something breaks.
>
>
> So IMO there is no validity saying that the prevalence of /24s in IPv4 private addressing indicates a preference for fixed subnet (prefix) lengths. IMO it is purely a case that the vast majority of people, even many you might think would know better, have never known anything else. Add in a sprinkling of devices which simply don’t work well in anything but a /24 (designed by people described in previous sentence) and it’s a self perpetuating situation.
>
>
>
> As to the topic in question. My suspicion is that even though some ISPs would look to monetise things - i.e. “pay us extra for a whole /64” - in practice I don’t think they’d survive too long in the same way that I haven’t heard for a while over on the ISC DHCP Users ML from anyone asking how to force periodic address changes*. Since it would be “a long time” before most devices worked with other than /64 prefixes, any ISP offering less would suffer from massive support problems (and hence costs) and probably get such a bad reputation that it either went bust or the bean counters realised it had been a massive mistake. In other words, the default /64 in IPv6 is probably deeply ingrained enough that the breakage by using anything less would prevent the otherwise inevitable race to the bottom.
>
> * Going back a few years, periodically there would be a question from a tech as some small ISP asking how to force clients to change IP address periodically - typically driven by an intension by the business bean counters to charge extra for a fixed address, and create pain for those wanting one but not paying. As I’m sure all of you know, that would go directly against the specs, and the common reply was to that effect, plus the ISC server didn’t support it, doing so would “annoy” users by breaking all their connections periodically, and further advising that if their employer still insisted to start looking for another job with someone who wasn’t going to be out of business in the near/mid term due to deliberately p’ing off it’s customers.
>
>
> My 2d worth, Simon
>
> P.S. I too started at the monkey see, monkey do school of learning. But have (through reading, and interactions with more experienced people like yourselves) learned better !
>
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [e-mail]
> List Info: https://mailman3.ietf.org/mailman3/lists/[e-mail]/
> --------------------------------------------------------------------
>