Re: [homenet] Let's make in-home ULA presence a MUST !?

[James Woodyatt <[e-mail]>]

On Fri, Oct 17, 2014 at 3:37 PM, Ted Lemon <[e-mail]> wrote:

> [...] before the network partitions, divide the ULA into 64k /64 prefixes,
> and distribute these evenly among attached routers.   Routers other than
> the ones that own a particular /64 are not allowed ever to use that /64
> unless the router that owns it relinquishes it to them explicitly.   Prior
> to partition, an agreement is made that one of the routers gets to keep the
> ULA in the event of a long-term partition.   When a partition happens, the
> routers that aren't able to reach that router anymore need to start a
> process of deprecating the ULA and adding a new ULA.   Presumably after
> something like half the total partition/abandon timeout period, they
> generate a new ULA and deprecate the old ULA.   If the network is
> reconnected after that, the new ULA is deprecated, the old ULA is
> advertised as preferred, and things heal.
>

My apologies, but this is the first I've seen this basic idea outlined
here.  If you explained it previously, then I missed it due to my
inattention, and I apologize.

I do think this idea should needs a more rigorous description before I can
give it more thoughtful consideration. Some questions: "an agreement is
made" how? "aren't able to reach that router anymore" why (and how is that
detected)? "the new ULA is deprecated, the old ULA is advertised as
preferred" which is which, and how is it recognized (e.g. by NTP time
stamp)?


> > p2. I also remain confused about the reasoning behind calling for a
> persistent locally-generated ULA prefix. In a previous message you said
> that it's okay for locally-generated ULA prefixes to expire, because there
> is no need for hosts on home networks to have any guarantee that at least
> one of its interface addresses is invariant over time, just that at least
> one of their addresses is never flash renumbered when a delegated prefix
> changes. As Lorenzo has demonstrated earlier today, this quality of never
> being flash-renumbered is easily met by delegated ULA and ordinary GUA
> prefixes.
>
> No, I actually said that it's worth making some effort to avoid having
> ULAs expire.   But it's impossible to eliminate situations in which they
> must.   My goal is simply to avoid having things like simple
> reconfigurations of the network topology result in the generation of new
> ULAs.


Okay... except it seems you're admitting that my scenario where a simple
reconfiguration of a network topology, e.g. one caused by an intermittent
RF interference on an unlicensed band of the radio spectrum, would result
in a fully regular and normalized generation of a ULA prefix that would
subsequently be deprecated on network rejoin and subsequently deprecated
again. This could happen several times per hour, right?


> > Returning to my question: why do we always need a locally-generated ULA
> prefix?  If it's to provide a time-invariant locally routable address to
> hosts, then locally generated ULA prefixes cannot ever be permitted to
> expire for any reason.  If they are ever allowed to expire, then they don't
> provide the time-invariant property.  However, if we don't actually need
> the time-invariant property, then what does a locally-generated ULA prefix
> do for us whenever one or more delegated prefixes is also present? It's not
> clear to me they are anything but absolutely redundant and unnecessary in
> that situation.
>
> This is a bit of a straw man (see previous comment).   I think trying to
> keep a permanent ULA is a good thing.   We can't always succeed, but we can
> make the set of circumstances under which we fail as small as possible.
>  This is in contrast to what you are proposing, which is that we
> essentially set out to fail, and see deprecation events whenever the
> upstream network goes down.


I certainly would agree with your observation that I don't see keeping a
permanent ULA as a good thing in itself. I keep asking why that would be
beneficial, and the answers continue to leave me scratching my head. Please
count me as one of the people here who has read Brian Carpenter's rant in
SIGCOMM about IP Addresses Considered Harmful, and who generally agrees.
I'm uncomfortable trying too hard to achieve some partial and dubious
success at establishing a persistent ULA prefix in the home network. I
think we should put our efforts more into making persistent the local
namespace for service registration and discovery.


-- 
james woodyatt <[e-mail]>
Nest Labs, Communications Engineering