Re: [v6ops] Two prefixes [draft-ietf-v6ops-icp-guidance WGLC]

[Arturo Servin <[e-mail]>]

	No, I didn't say that.

	1) if you are small, multi-homed you should use BGP+PI.


	2) if you are small and uni-home use PA.

		
	If 1 does not scale (I would question that), it is a problem that we should solve elsewhere.

	There are other solutions for 1, but I would not recommend them as a best practice, just like workarounds to use PA and avoid BGP.

	1 and 2 are orthogonal to the fact that the ICP is as well an enterprise or not.

Regards,
as

On 8 Aug 2012, at 12:16, Brian E Carpenter wrote:

> On 08/08/2012 15:59, Arturo Servin wrote:
>> Brian,
>> 
>> 	But we are not talking about enterprises, we are talking about ICPs, which IMHO are two different problems and advices.
> 
> To be clear, I intended to talk about either pure specialised ICPs, or an enterprise that
> wishes to behave as an ICP for its own content.
> 
>> 	For Enterprises, yes PA and PI are debatable (as discussed in many places, many times) and more dependent on the size and type of enterprise, for ICPs I think we are more or less in consensus that PI is more much common (and IMO the recommendation).
>> 
>> 	Perhaps, the corner case are "small ICPs" which if they are uni-homed, then PA is fine, if they are multi-homed then BGP+PI should be the recommendation no matter if they are small or large.
> 
> But that's the case that doesn't scale to millions. Are you suggesting that
> we have to tell them "you can't serve your own content reliably, because
> you're too small"?
> 
>    Brian
> 
>> Regards,
>> as
>> 
>> On 8 Aug 2012, at 09:12, Brian E Carpenter wrote:
>> 
>>> My concern here, as I've already hinted, is publishing advice that is
>>> valid for the N thousand enterprises that can reasonably expect to get
>>> a PI prefix but nothing for the N million others that may wish to serve
>>> up content *and* be multihomed.
>> 
>>