Re: [v6ops] Calling the question: draft-ietf-v6ops-host-addr-availability-03.txt

["Templin, Fred L" <[e-mail]>]

Hi Philip,

> -----Original Message-----
> From: v6ops [mailto:[e-mail]] On Behalf Of Philip Homburg
> Sent: Monday, December 21, 2015 12:08 PM
> To: [e-mail]
> Cc: [e-mail]
> Subject: Re: [v6ops] Calling the question: draft-ietf-v6ops-host-addr-availability-03.txt
> 
> >I feel very strongly that we should not do document what hosts need to do
> >in this document. That is out of scope - what we're saying here is
> >providing recommendations on how networks should act, not how hosts should
> >act.
> 
> Maybe should you take it one step further and cut out all text on how to
> assign multiple addresses.
> 
> For me there is no doubt that networks should provide hosts with
> multiple addresses. That is clearly BCP.
> 
> >Like I said: on networks that require explicit requests for address space,
> >we simply have no other way to assign anything that's not a handful of
> >addresses. By definition, on those networks SLAAC is not supported, so all
> >we have is IA_NA. So that's the B in BCP, though it's not really so much
> >"Best" as "Only".
> 
> So, we don't know. It might work. Nobody seems to have any experience.
> So cut the speculation and say that we don't know.
> 
> (Alternatively, reclassify and informational and the current text is fine)
> 
> >We do have plenty of experience with using DHCPv6 PD to assign addresses to
> >things, and yes, those things include hosts (e.g., Windows, which has
> >supported DHCPv6 PD for close to a decade). It's true that not all hosts
> >and networks support this, just like not all networks support DHCPv6 or
> >SLAAC or whatever else. But some enterprise networks do. Our enterprise
> >network has places on it that do PD, and as you might expect, the servers
> >that hand out the prefixes neither know nor care if it's a host or a router
> >that requested them. And there are DHCPv6 PD implementations for host OSes
> >such as ISC DHCP and dibbler.
> 
> If people have been doing PD for Windows hosts for close to a decade,
> find somebody who has that kind of experience and describe how it
> works in practice. (Maybe in an informational draft)
> 
> >If your concern is that using DHCPv6 PD on a host will cause lots of
> >problems - what problems are you talking about? From the host perspective
> >it's simple. You get a prefix.
> 
> I described it earlier:
> - A host gets a prefix from DHCPv6 relay. The relay reboots, does the
>   relay have a persistent store of the PD leases or is the host expected
>   to poll?
> 
>   As far as I know, there is no text in RFC 3633 that requires persistent state
>   in a DHCPv6 relay. Maybe I missed something.
> 
>   If the host is expect to poll, how does that work. You mention Windows,
>   how do they deal with this problem.
> 
> - And to spice it up, how do redundant routers, acting as DHCPv6 relays and
>   one of them is shutdown. Again is this something the network has
>   to handle, or is this the resposibility of the host. It is just not
>   documented. If PD for hosts is used in practice, then somebody must have
>   dealt with this problem.
> 
> I don't see how this is 'simple'.

I thought I had already answered this, but for AERO there may be many routers on
the link. The client discovers the unicast addresses of routers via a name service
lookup and maps the multicast group "All_DHCP_Relay_Agents_and_Servers"
to the unicast addresses discovered in the name service. By this, I mean that 
through encapsulation the inner destination address is multicast and the outer
destination address is the unicast address of a router.

The client then does a DHCPv6 PD Solicit/Request through each router it wants to
associate with. Each router in turn configures both a DHCPv6 Relay (via RFC6221
Lightweight DHCPv6 Relay Agent) and a DHCPv6 Server. When the client requests
a DHCPv6 Prefix Delegation, it will get the *same* prefix(es) from all routers it
associates with. Meaning that all routers have access to the same database of
prefix-to-client mappings, and all routers will provide an identical service.

When one or more routers go down, the network will notice this and withdraw
the prefixes they are advertising from the routing system. The client will also
notice that the router has gone down and associate with a different router
or routers. The routing system will then direct traffic destined to the client
via one of the new routers.

This works fine even if the client is mobile and continuously moving from
access network to access network. And, the client gets the same prefix(es)
every time it attaches to the network so there is no renumbering. But,
most importantly, it gets the same service from whichever router(s)
it chooses to associate with, so it can even move from router to router
if needs be to reduce path stretch.

This is all in the AERO spec. I think an informative reference in this document
would be in order.

Thanks - Fred
[e-mail]

> 
> _______________________________________________
> v6ops mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/v6ops