Re: [69ATTENDEES] DHCP

[itojun@itojun.org (Jun-ichiro itojun Hagino)]

> > 	RA-only and DHCP-only houses there will surely be. They may be
> > 	a majority; I just don't know.  Vendors and implementors will
> > 	no doubt be happy to accommodate both of these, as they can offer
> > 	a complete solution.  There will also be houses where neither of
> > 	these quite meets the requirements (technical, layers 8 and above,
> > 	or whatever).  These houses will need to profile their RA and DHCP
> > 	services so that they "play nice" with each other.  They will look
> > 	to vendors and implementors who will have taken care to allow
> > 	different slices of either service to be disabled so that it can be
> > 	offered only by the other service.
> > 
> > 	One size just won't fit all.
> > 
> > 	Stop me if I haven't noticed that this is all just a "violent  
> > agreement".
> 
> Well, I fully agree with you.  "Networks differ, so do requirements".
> 
> I have some issues with some folks from the DHCP camp refusing to understand
> that there are operators that just follow a different operational model.

	i agree that vendors would implement whatever they would need to, but
	iirc the group is here to give guidance for operations.

	the primary issue with DHCP-based address assignment is the requirement
	for DHCP server to keep state.  due to this, major DHCP server
	implementation suffers from reading in lease database when it reboots
	(stops for 5 minutes).  DHCP server is a single point of failure.

	with RA-based address autoconfiguration and DHCPv6-lite based stateless
	DNS server address passing, there's no need for keeping state in the
	server side, hence there's no single point of failure.

itojun