[c-nsp] IPV6 in general was Re: Large networks

[Phil Mayers]

Gert Doering wrote:
> Hi,
> 
> On Thu, Aug 27, 2009 at 09:35:54AM +0100, Phil Mayers wrote:
>> IPv6 emulated the then-state-of-the-art IPX autoconfig mechanisms, and 
>> seems reluctant to admit it's missed out the last decade of operational 
>> knowledge acquired with IPv4.
>>
>> SLAAC should die the death it so richly deserves (except for link-local) 
>> and DHCPv6 should gain prefix advertisment capability.
> 
> SLAAC works *very* well for the things it was made for: zero-conf 
> environments, with no dedicated DHCP server - as in "home networks" or
> "office networks".

Hmm.

It seems to me that there are two types of networks:

  1. Not connected to the internet at all i.e. no router etc. - in which 
case SLAAC can just allocate link-local addresses

  2. A small network with a router - in which case, I fail to see why 
the router can't embed a DHCPv6 server just as easily as 
SEND/CGA/RFC5006 RA implementation

Basically, I think SLAAC should only ever allocate link-local. I'm fine 
with that - it does, as you say, do the job very well, and the ability 
to bootstap subnet-local connectivity off link-local makes the next 
steps much easier, cleaner and saner.


> 
> It's not meant to be used for connecting customer sites to an ISP 
> network, and it's not overly useful for numbering servers either - but
> that doesn't make it "death deserving".

Fair point.

The problem is that there seems to be a zealous effort to shoehorn 
everything into SLAAC (e.g. DNS servers) and avoid bringing any "IPv4 
mistakes" into IPv6.

But some people seem to think DHCP is a "mistake", and DHCP options a 
"mistake" and allocating fixed IPs a "mistake". I cannot share that view.

I think that holds IPv6 back, because a lot of enterprises aren't 
willing to be the guinea pigs for an unproven model where potentially 
rogue clients generate their own addresses as they please.

I do think that there's nothing SLAAC can do that DHCPv6 can't do, if a 
prefix advertisment DHCPv6 option were created. And as I say, I wonder 
therefore what the point is - why couldn't these tiny home & office 
networks with ADSL just as easily embed a minimal DHCPv6 server, versus 
an RFC 5006 implementation?

Ironically we can't use DHCPv6 here even if we wanted to, because it 
doesn't work in 6vPE under SXI - I was told it was "harder than it 
seemed" and my TAC case was closed as the opened a PER - I think...