On Oct 7, 2011, at 6:08 PM, Erik Nordmark wrote:

> On 10/7/11 2:18 PM, Ole Troan wrote:
>>>> I think it's
>>>> reasonable to write to stable storage whenever you configure a new
>>>> prefix or change an existing prefix (but not if you simply refresh it).
>>>> If the router comes back up and nobody has the prefix, it would simply
>>>> claim it back. If someone else has it, it would have to just form a new
>>>> one using loop detection.
>>> 
>>> In your view, if there is a single prefix assigned to the home by the ISP, 
>>> and there are three routers connected to a particular link in the home, 
>>> would each of those three routers assign a different prefix to the link?
>>> 
>>> Am I missing something?
>> 
>> see the zospf draft. the DR of the link will assign the prefix for the link. 
>> the BDR should also store that chosen prefix into stable storage. of course 
>> if you power off both the DR and the BDR and power up a new 3rd router, the 
>> link will get a new prefix.
> 
> The zospf draft (see section 3.6) declares a conflict if two routers attached 
> to the same link pick the same IP subnet prefix for that link. That doesn't 
> seem what we want.
> 
> And zospf doesn't talk about using stable storage; doesn't seem to have any 
> notion of providing stable prefixes assigned to the links.
> 
> Thus this problem remains to be solved AFAICT.

> 
> I think we first need to agree on what we think are the requirements on 
> prefix stability. What if one of the routers attached to the link fails and 
> is replaced by a different router?
> What if a link partitions? And later merges?
> What the user intentionally splits (partitions) a link (e.g., to create a 
> separate link for the teenage game room); should the network automatically 
> pick a new prefix for that new link? (How does that interact with the prefix 
> being in stable storage on the router.)
> 
> One thing to keep in mind as we think about these requirements is that if the 
> user has created a bridged Ethernet with no routers, all of the above will 
> work naturally; might take seconds to relearn when the topology changes but 
> there is no manual intervention needed.
> Is that the goal for this prefix assignment?

No manual intervention is definitely a goal.

> 
> Who is working on the requirements document? ;-)

For now, the arch document will be capturing as close as we are getting to 
formal requirements in an ID.... mostly principles and high-level requirements. 

I'm not convinced a detailed or separate requirements document for any of our 5 
areas could be produced without it becoming a reverse engineering exercise from 
various solutions anyway. We tried to have a formal "just the requirements" 
discussion during the meeting, and what I heard was quite a bit of solution 
discussion couched in requirements language anyway.

- Mark

> 
>   Erik
> _______________________________________________
> homenet mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/homenet

_______________________________________________
homenet mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/homenet

Reply via email to