RE: draft-bourbaki-6man-classless-ipv6-00

["Manfredi, Albert E" <[e-mail]>]

From: Mark Smith [mailto:[e-mail]] 

> * DHCPv6 does not record addresses in use on a link. It will not
> record link local addresses or manually configured addresses. It
> is only recording hosts that asked to acquire addresses via DHCP.

Okay, like DHCPv4. But when the system admin needs stable addresses for hosts in the network, e.g. a peer-peer network that does not use a DNS, DHCPv6 PD is a good tool to have.

> * You're overlooking the cost and service impact of renumbering to
> increase the size of the subnet if it isn't large enough. That can
> be a large cost and large impact if many hosts are involved. So
> large, I've seen people avoid paying it when there were in the
> order of 1000 hosts involved on an IPv4 network - the hosts were
> spread across 4 x /24s on the same Ethernet segment.

Depends on the scenario. In my reality, expanding a network at the edges easily, without involving in any way the authority that allocates blocks of addresses, is quite valuable. A scheme that makes this difficult is instead a nuisance.

What's behind this debate, from my point of view, is a need for the same functionality offered by NAT in IPv4, but without incurring the penalties of using NAT. A fixed IID length of 64 bits, when /64s or maybe even /56s are what customers are given, is a bit like never having invented NAT for IPv4. It's limiting. I realize this view is not everyone's concern.

Bert