Re: A common problem with SLAAC in "renumbering" scenarios

[Fernando Gont <[e-mail]>]

Mark,

On 20/2/19 03:16, Mark Smith wrote:
> On Wed, 20 Feb 2019 at 15:22, Fernando Gont <[e-mail]> wrote:
>>
>> On 20/2/19 00:11, Ole Troan wrote:
>>>
[....]
>>
>>> With regards to the addressing model, your question shows a certain lack of history, but given that we all gave lived under the reigns of NAT for so long.
>>
>> You have been referring to an addressing model that implies
>> static/stable addressing. I simply asked where that model is described.
>>
>> If such requirement is not spelled out somewhere, I don't know why you'd
>> refer to that as a model, or even why you could expect it.
>>
>> Not sure how my question for a reference that backs your model can turn
>> into something I'm supposed to be lacking, missing, or ignoring.
> 
> I think it is implicit in a lot of places, if not stated explicitly
> anywhere (although it may be somewhere in an RFC like RFC1122).
> 
> - By default, hosts don't delete addresses from interfaces if the
> interface goes down. Addresses persist across interface down/up
> events. If the most common cause of hosts' interfaces going down and
> coming back up was considered to be attachment to a different link,
> then addresses would probably be deleted by default upon link down.
> 
> - As Ole implied, TCP uses addresses as identifiers, so TCP expects
> that the address will persist for the entire duration of the TCP
> connection, regardless of how long that duration is.

Your are turning a kludge into a design principle. Yes, TCP uses
addresses as identifiers, because the Internet doesn't have a
name/identifier for hosts.

Then instead of creating such identifiers, folks tried to "move
addresses" -- well, if you can move it, it's not an address anymore!



> - TCP connections survive the host's local interface down/up events.

Not necessarily. This is implementation dependent.

In fact, if this was the case, and with the current lifetimes, with
Valid Lifetimes of one month, imagine what your network config would
look like after moving around a few coffee shops in a week -- unusable!


> - The SLAAC default preferred and valid lifetimes are very exceptional
> and remarkably large compared to other typical default networking
> timer values that are in the order of 1s, 10s or 100s of seconds. The
> default Valid Lifetime is ~2.6 million seconds/30 days.
> 
> Considering that a single RA would reset a decrementing address's
> Valid Lifetime back to ~2.6 million seconds, I think that suggests
> that TCP can expect that an address it is using on a host can be
> relied upon to exist for a minimum of 30 days, so a TCP connection
> should be able to be relied upon for a minimum of 30 days. That's
> exceptionally robust given that the unsolicited RA default maximum
> interval is 600 seconds.

Clearly, with your model of TCP connections surviving interface up/downs
and the default lifetimes, you really have a host that is unusable,
except in very specific network conditions -- such as: the host never
moves (quite unlikely these days) and nothing (including the CPE) fails
(quite unlikely).

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