Re: RA Option Guidelines

[Mark ZZZ Smith <[e-mail]>]

Hi Lorenzo,


________________________________
From: Lorenzo Colitti <[e-mail]>
To: Mark ZZZ Smith <[e-mail]> 
Cc: Brian E Carpenter <[e-mail]>; Hemant Singh (shemant) <[e-mail]>; "STARK, BARBARA H" <[e-mail]>; Brian Haberman <[e-mail]>; "[e-mail]" <[e-mail]> 
Sent: Tuesday, 7 April 2015, 10:29
Subject: Re: RA Option Guidelines



Bear in mind that AIUI there is pretty strong consensus in the dhc working group that DHCP should *not* be used for general application configuration, but only for bootstrapping networking parameters.


/ I'm not sure I understand their view, although if 'network parameters' has been generalised to include application parameters then it makes more sense.

/ I'm considering "network parameters" to mean only network layer/layer 3 parameters. Other parameters belong to other upper layers, and those other layers are only implemented in hosts (or the (control plane located) host component/functions of a router), so they're host oriented rather than network parameters. Going by the name, it sounds like the Dynamic Host Configuration Protocol is either the best or a likely candidate protocol to configure these transport layer+ parameters.
/ RFC5505, "Principles of Internet Host Configuration" also creates a configuration distinction between the network layer ("Internet-Layer Configuration") and the upper layers ("Higher-Layer Configuration"), and also discusses the issue of DHCPv6 learned parameters been less dynamic in "Relationship between IP Configuration and Service Discovery".
/ Regards,Mark.

On Tue, Apr 7, 2015 at 8:20 AM, Mark ZZZ Smith <[e-mail]> wrote:

So that sounds to me like a limitation in DHCPv6 that should be fixed, rather than trying to work around it with RAs.
>
>
>I think the fundamental question about what to use RAs for comes to:
>
>
>Do we bind/tightly couple the ability to deploy new applications to the features supported by the local router or not?
>
>
>
>My understanding of the Internet architecture as described in RFC1958 is we should not.
>
>
>________________________________
> From: Lorenzo Colitti <[e-mail]>
>To: Brian E Carpenter <[e-mail]> 
>Cc: Hemant Singh (shemant) <[e-mail]>; "STARK, BARBARA H" <[e-mail]>; Mark ZZZ Smith <[e-mail]>; Brian Haberman <[e-mail]>; "[e-mail]" <[e-mail]> 
>Sent: Saturday, 4 April 2015, 23:26
>Subject: Re: RA Option Guidelines
> 
>
>
>
>
>
>On Sat, Apr 4, 2015 at 4:16 AM, Brian E Carpenter <[e-mail]> wrote:
>
>Do we actually agree on this, or rather on a slightly expanded version:
>>
>>RA options should be limited to link-local/subnet specific parameters
>>(i.e., layer 3 parameters only) necessary for a simple unmanaged network.
>>The only exception is RFC 6106 (DNS Configuration). More complex situations
>>are better handled by other configuration metods such as DHCPv6.
>
>
>No, I strongly disagree.
>
>
>There are other differences between RA and DHCP. The most important one I see is that with DHCP it is basically impossible to change things dynamically.
>
>
>Dynamically changing stateful parameters requires that the client implement reconfigure, and even then, it requires authentication, and requires either that the DHCP server never crashes, or that DHCPv6 failover is in use. Dynamically changing stateless options doesn't even seem possible at all - you have to wait until the client asks again, and the minimum time that must pass is 10 minutes (RFC 4242).
>
>