Why not RIO? (Re: Geoff on IPv4 Exhaustion)

[mdupuis at gmail.com]

Another pragmatic example that may be helpful would be in virtualized hypervisor clusters. Large collections of hardware on a common management segment typically are managed on a MAC address basis via dhcp. The hardware id (MAC) is programmically modified via the dhcpd configuration to allow hardware to receive specific configuration information based on the allocated role. Modification of the hardware role involves evacuation of guests off the hypervisor, reboot, and receipt of new network assignment.  

Depending of the hardware role assignment, support systems such as syslog are typically reachable via alternate gateways depending on traffic volume, lawful access requirements, or a host of other requirements. 

-m

 
Sent from my BlackBerry device

-----Original Message-----
From: Phil Mayers <p.mayers at imperial.ac.uk>
Sender: ipv6-ops-bounces+mdupuis=gmail.com at lists.cluenet.de
Date: Sun, 20 Nov 2011 12:43:18 
To: <ipv6-ops at lists.cluenet.de>
Subject: Re: Why not RIO? (Re: Geoff on IPv4 Exhaustion)

On 11/20/2011 04:23 AM, Brian E Carpenter wrote:

> When you write "hosts'" do you mean that you need to configure
> each individual host's routing information *separately*?
>
> If the answer is yes, what is the reason use case for needing to configure
> each host separately?

Student machines route via gateway X. Staff machines via gateway Y. Or 
registered machines via router X, unregistered machines via captive 
portal Y. Or machines with IP addresses from the newly acquired IPv6 
range via router X, machines with IP addresses from the old range via 
router Y (think mergers & acquisitions). Replace these categories with 
anything else that makes sense, I can think of a few more examples.

Registered machines use DNS servers X, unregistered machines use DNS 
servers Y. Repeat as above.

RA for gateway/DNS/address config with the client "deciding" which to 
use makes these problematic at the very least.

I am not saying this kind of solution is good or desirable - I hold no 
strong opinion - but it seems clear to me from this thread and lots of 
"hate IPv6 RA" discussions that I've heard in other places and at other 
times, that people want to control the router and other settings on a 
per-host basis.

At the very least, this seems like it would be a useful tool to have in 
the toolbox.

Frankly, I can understand why, if you want to do that and have been 
doing so for years in IPv4 with DHCP, the sorts of purist responses some 
people make:

  - "use ICMP redirects!",
  - "then you should be using a routing protocol!"
  - "the internet sees blah blah as damage and routes around it!"

...are not just unhelpful, but jaw-clenchingly irritating.

Can anyone who spends any time at the IETF give some insight into why 
exactly DHCPv6 is the poor cousin of RA?

Cheers,
Phil