SLAAC(autoconfig) vs DHCPv6

Mon Aug 18 21:11:16 UTC 2008

On 18 aug 2008, at 22:23, Dale W. Carder wrote:

> - really, really, really broken: it didn't support handing out
>  any DNS info until RFC 5006, thus SLAAC still requires human
>  intervention on a client to make "teh v6 interwebs" work.

While I agree that it is bad that the DNS configuration issue took so  
long to fix, I wouldn't consider this a flaw of stateless  
autoconfiguration, which works extremely well. There have been many  
times that I was at conferences where the IPv4 DHCP wouldn't work so  
it was impossible to go online, while stateless autoconfig rarely  
creates any problems. (Although there could be connectivity problems  
upstream, though.)

> DHCPv6
> - doesn't ship w/ some OS's

Forget about it on XP, but it's in Vista. You can add it to BSD/Linux  
without too much trouble (are there good, bugfree implementations for  
those yet?) but Mac is a problem for prospective DHCPv6 users because  
the network configuration mechanisms are fairly proprietary and DHCPv6  
isn't likely to be supported any time soon.

> - new (danger code), not all features implemented
> - router support for dhcpv6 relay very limited
> - advanced things like prefix delegation don't really seem to
>  have been ironed out.

Actually the prefix delegation has worked just fine for me. This is  
the redeeming feature in DHCPv6.

In my opinion, DHCPv6 was severely misdesigned. For instance, there  
are stateful and stateless variations, and the _client_ has to choose  
which to use. DHCPv6 also doesn't give you a subnet prefix length or a  
default gateway, so you still need router advertisements (that are  
also used for stateless autoconfig). The latter can be considered a  
feature, but I'm guessing the lack of a subnet prefix other than the  
assumption that the whole world uses /64 has been giving DHCPv6 server  
implementers a lot of headaches.

> In case you weren't confused enough between the two, they are not
> mutually exclusive.  You can run both SLAAC and DHCPv6 at the same
> time on the same L2.

Of course there's no telling what exactly the clients are going to do  
in that case...

Iljitsch