IPv6 multihoming

[Brian E Carpenter]

On 2011-02-06 05:10, Henrik Lund Kramshøj wrote:
> On 04/02/2011, at 23.52, Brian E Carpenter wrote:
> 
>> On 2011-02-05 10:13, Nick Hilliard wrote:
>>> On 04/02/2011 19:31, Brian E Carpenter wrote:
>>>> If you mean a user site running two different PA prefixes all the
>>>> time, so that no holes need to be punched, that works out of the box,
>>>> although people don't seem to believe it. It also scales.
>>> yep, it scales well.  It doesn't actually do what end users actually
>>> need from multihoming, mind you - but it does scale very nicely indeed.
>> It doesn't support session failover unless both ends run shim6.
>> And there is still some missing technology for exit router selection.
>> Certainly, we need to solve those problems. My point is that we need
>> to, since the PI approach still won't scale (as we've known since
>> about 1992, btw). It will do fine for a year or five.
> 
> and running servers is straight forward and updating DNS dynamically when
> one of the connections fail are easy, 

I don't see the need to do that. In any case the DNS TTL will mean
that it doesn't really do much to help.

and getting to influence the path coming into your network
> can be done without problems ....

Clearly the traffic sent to the prefix belonging to each of your
ISPs will arrive via that ISP. So source address selection controls
the return path.

> 
> I guess we are really talking about different things, 
> 
> yes a user browsing and USING ressources on the internet can use multiple roads out of the network
> 
> but from my viewpoint having servers and services running is a whole other matter.
> 
> We are hosting a number of really large accounts and we have enough complexity when providing redundant services, using multiple locations
> using each their own addresses (some from PA some from PI IPv4) that adding multiple
> addresses into the mix from TWO (or more PA spaces) is really not an option.

And in that situation, maybe PI is the correct answer. You are part of the thousands,
not part of the millions, when we discuss scaling.

> 
> 
> We have streaming servers and would like to peer traffic away, as well as having two or
> more redundant transit providers.
> 
> Do you, Brian, actually suggest that this service would have to have ~5 addresses
> from the various PA spaces? instead of having a well-known IPv6 PI address which is
> the same from various places?
> 
> Multihoming and IPv6 PI MUST be sorted out, and fixed - not in 5 years, but more like now!

PI is what it is, and I don't think it really needs sorting out as long as we
can limit it to thousands rather than millions of sites. It's multihoming
for the millions that is a very hard problem.

> I wrote my thesis about IPv6 in 2002 and even then people could not really agree how to do multihoming, *sigh*

Because it's really hard; I mean mathematically hard. I won't waste time here
by listing the currently active solution proposals.

> 
> Some people discuss on mailing list, others try to work with the internet and provide services :-)
> 
> Hope the message gets through, no intention to annoy anyone, just a bit frustrated that IPv6 is
> finally getting off the ground.

I fully agree; it's very frustrating.

   Brian

> Best regards
> 
> Henrik
> 
> 
> 
> --
> Henrik Lund Kramshøj, Follower of the Great Way of Unix
> hlk at kramse.org hlk at solidonetworks.com 
> +45 2026 6000 cand.scient CISSP CEH
> http://solidonetworks.com/ Network Security is a business enabler
> 
> 
> 
> 
>