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

On 8/7/12 9:06 PM, Cameron Byrne wrote:
> On Tue, Aug 7, 2012 at 6:42 AM, Brian E Carpenter
> <brian.e.carpenter@gmail.com> wrote:
>> I think that's unfair and kind of ignores 
>> draft-v6ops-multihoming-without-ipv6nat It works today. There are 
>> known difficulties with address selection and with ingress filtering, 
>> of course. And it's a bit more fiddly to configure routing and DNS 
>> for IT crews used to the old way of doing things. But it really isn't 
>> unknown territory.

I think characterizing the criticism as unfair is unfair. As far as I'm 
aware nobody actually serves from two PA prefixes from the same hosts in 
practice... While DNS GLB is suitable for doing that, what it isn't 
suitable for is optimizing provider exit selection, which BGP is rather 
good for and which proper BCP 38 RPF checks prevent unless I advertise 
the respective prefixes to my upstream...

So either my hosting provider needs to be multihomed using pi or 
delegates PA that others are willing to carry, or I need to be... and 
what I use the GLB for is to direct traffic to one instance of a service 
or another.

finding a multi-homed hosting provider in ipv4 land isn't that hard it 
shouldn't be in IPv6 either.
>>> and therefore should not be a recommended approach. I
>> If we are only addressing a few thousand sites, sure, but how else do
>> you suggest we deal with content providers by the million?
>>
>>> understand ipv6 was designed to work this way. .... But afaik, it is not
>>> really exercised.  If someone has done it in a production network, that
>>> would be good to know
>> Yes, facts are always good.
>>
>>      Brian
>>
>
> One of the keys here that i overlooked is the ICP will certainly not
> be using SLAAC.  They will be using static addresses manual, automated
> manually (Puppet, ...) , or via DHCPv6.  That said, this is an issue
> of automation and instrumentation on how addresses are assigned and
> therefore should not be much of challenge, including running multiple
> prefixs in a short transition period.
>
> Meaning, this problem is solved in IPv4 for ICPs and those same IPv4
> solution apply in IPv6.
>
> CB
> _______________________________________________
> v6ops mailing list
> v6ops@ietf.org
> https://www.ietf.org/mailman/listinfo/v6ops
>