Re: [v6ops] New Version Notification for draft-yourtchenko-ra-dhcpv6-comparison-00.txt (fwd)

[Nick Hilliard <[e-mail]>]

On 19/12/2013 01:34, Lorenzo Colitti wrote:
> I understand that you think that opinion is not a valid reason, but I would
> like to point out that yours is also an opinion. Yes, DHCPv6 allows you to
> configure more options than RAs, but it's also less flexible in some ways
> (e.g., doesn't support load-balancing over multiple routers, doesn't easily
> support deprecation, etc.). So I think it's clear that one is not
> necessarily better than the other.

I'm not claiming that RAs are better than DHCPv6 or the other way around.
If you want to run only SLAAC or benefit from any of the features of
RAs+SLAAC, then that's great and I'm very happy for you.

I'm just stating that for my deployment scenarios, I specifically do not
want the extra stuff that RAs/SLAAC provide like load-balancing over
multiple routers, deprecation of defgws, multiple networks/gateways per l2
domain and all that.

On the other hand, I need to run DHCPv6 because RAs do not provide me with
the technical knobs that I need.  These include address assignment (rather
than slaac), and other dhcp-only knobs.

In this context, RAs serve only to provide defgw and (as Tore kindly
reminded me) prefix assignment, and then hand everything off to DHCPv6.

Leaving aside the operational issues associated with running separate
configuration on separate systems for the same end goal of
autoconfiguration, from a protocol simplicity point of view, this approach
is not very sensible because, we have an entire protocol dedicated to
handling something that with a modest amount of effort could probably be
handled by dhcpv6.

I'm sorry you feel this is a dead horse.  My take is that if we're
designing a protocol which has an expected lifetime of $manyyears, then
protocol simplicity should be high on the list.  Given the amount of recent
noise on this list discussing the interaction between RA+DHCP, it's
abundantly clear that simplicity has been lost in favour of confusion.
Simplicity is a critically important design goal for anything related to
engineering, and the old maxim is true: "everything should be kept as
simple as possible, but no simpler".

If possible, I'd really like if we could have a non-heated discussion about
the technical pros and cons of whether standalone DHCPv6 is viable.

I know that for my use cases, it would be a highly desirable endpoint, but
I understand that other people have different deployment requirements.

Nick