[gPXE] Thoughts on an RFC draft pertinent to DHCPv6 netboot

[Stefan Hajnoczi]

On Mon, Jun 28, 2010 at 2:48 PM, Jarrod Johnson
<jarrod.b.johnson+gpxe at gmail.com> wrote:
> I was interested on people's thoughts on the IETF draft
> at http://tools.ietf.org/html/draft-narten-dhc-duid-uuid-00.  The idea is to
> have firmware use UUID from the DMI tables as the DUID so that the operating
> system can reliably select the same DUID as the NIC (if DUID-LLT, then the
> timestamp will almost certainly never line up, if DUID-LL, it may be
> workable for single-NIC systems, but multi-NIC boxes are in a fix).

Thanks for posting this draft.  Here are some thoughts to check I have
understood this correctly:

The problem addressed by DUID-UUID is persistent machine
identification for DHCPv6.  The existing DUID types are problematic
because they make it difficult for DHCP clients in multiple software
components (netboot firmware, bootloader, OS) to use a consistent DUID
across all software components.

The DUID-UUID is a per-system identifier that is universally unique.
On x86 the UUID can be taken from the DMI firmware tables.  The draft
document is not clear on which exact value from the DMI tables is
chosen.  Perhaps it is obvious (I don't know DMI) so I looked at
dmidecode output on my machine:

Handle 0x0001, DMI type 1, 27 bytes
System Information
    UUID: ...

Do you want to explicitly state that this DMI field is the one to use?

Stefan