Re: draft-ietf-v6ops-addcon-03.txt ... ULAs of shorter-than-/48 and ULA multicast scope matching ...

[Mark Smith <ipng@69706e6720323030352d30312d31340a.nosense.org>]

Hi Tim,

On Tue, 13 Mar 2007 09:41:12 +0000
Tim Chown <tjc@ecs.soton.ac.uk> wrote:

> On Tue, Mar 13, 2007 at 07:30:51PM +1030, Mark Smith wrote:
> > 
> > I'm slightly confused be this, and I think the comment about not
> > being able to summarise 16 /48 ULAs is what is confusing me.
> > Summarisation seems to me to only be useful if you are going to
> > announce a summary to an upstream entity of some sort, and is really
> > only for the benefit of the upstream entity (e.g. upstream ISP, IGP
> > backbone area, other areas carrying inter-area routes rather than a
> > default etc.), as it reduces their route table and other routing
> > resource requirements.
> > 
> > As ULAs are kept within private entity, and therefore there isn't any
> > upstream entity, is there really any need to summarise to less than
> > 16 /48 ULAs (in your scenario, for example)? If, within your
> > organisation, your routers can't cope with 16 /48 ULAs, I'd think you
> > problably have bigger routing problems to deal with (and a default
> > route only might be a main (and only) solution to that)!
> 
> Well, were we talking of maybe a /36 rather than a /34, with aggregated
> internal routing for sites within the prefix, then I think the above
> approach could be useful.
> 
> > If you were summarising ULAs internally to scale an IGP by dividing the
> > IGP into areas, would 16 areas, corresponding to the 16 /48s (global
> > and ULAs), be too many to cope with ? I'm not really sure they would.
> 
> Well, think a little bigger than a /44 example?   What would you say then?
> 

We'll, I've understood that a /48 was sized to be big enough for
nearly everybody, and those who require larger would only maybe need
a /47 or a /46. A /44 seems extremely large to me (around 1M subnets),
so I'm wondering whether going much shorter is that practical - unless
we might be talking about a service provider, who might have 1M+
distinct subnets, who then maybe would want to have fully matching ULA
subnets. If they're that size, I'd be assuming they'd be aggregating
internally fairly aggressively for both global and ULAs, just to cope
with router limitations. Possibly they might not deploy ULAs for all of
those subnets either - without putting much thought into it, I don't
think there would be that much benefit in assigning ULA subnets to
their customer attachment network edge links. (Possibly SPs may not
deploy ULAs at all, or at least only for their internal corporate /
support networks, not the customer traffic carrying links.)

> > I do see the operational benefits of having the /44 to /64 bits in the
> > global and ULA addresses match. I'm not quite sure I see how
> > summarisation by itself, which is how I read your paragraph, would be a
> > justification to do that, when the consequences are increased chances
> > of collision, should you join your /44 ULA domain with somebody else's.
>  
> Aligning the bits in the global and ULA addressing plan may have advantages
> as you say.   That's the other aspect.
> 
> My personal concern is that network admins will pick all zeroes for the
> ULA prefix, because they would rather have easy to type addresses than
> a potential benefit for leakage/merger/etc.
> 
> > Have I somehow misunderstood what you're trying to get at?
> 
> No, your question is a good one, and we need some discussion :)
> 

Regards,
Mark.