Syslinux - May 2007 - Infiniba(n)d and Firewire (1394)

I'm currently looking at how to deal with the lack of usable MACs in
Infiniba(n)d and Firewire.  Instead of presenting the MAC address when
querying for the config file, these guys need either the Client
Identifier, or the PC UUID.  The Client Identifier is a fairly opaque
object which tends to include various fixed portions; the PC UUID is far
to often not at all or incorrectly programmed.

I'm interested in how people who have DHCP setups with IB or 1394 handle
this, and also how long the Client Identifier really is (in bytes).  My
first plan was to simply expand it out in dash-separated hexadecimal the
way I currently do with the mac, but such a filename could get *very*
long which can cause a whole lot of trouble.

	-hpa

H. Peter Anvin wrote:
> I'm currently looking at how to deal with the lack of usable MACs in
> Infiniba(n)d and Firewire.  Instead of presenting the MAC address when
> querying for the config file, these guys need either the Client
> Identifier, or the PC UUID.  The Client Identifier is a fairly opaque
> object which tends to include various fixed portions; the PC UUID is far
> to often not at all or incorrectly programmed.
> 
> I'm interested in how people who have DHCP setups with IB or 1394
handle
> this, and also how long the Client Identifier really is (in bytes).  My
> first plan was to simply expand it out in dash-separated hexadecimal the
> way I currently do with the mac, but such a filename could get *very*
> long which can cause a whole lot of trouble.
> 
One option that I've considered is to still hexify it, but do it without
dashes (except for the leading type byte), so an Infinibad example might
potentially look like:

pxelinux.cfg/20-ffa9d5fe460003c206a4fbe25fec4e9e15f96818711b2da5de9fca

I'm not sure if anyone could even recognize what that was, or if that's
really the form Infiniband nodes use for their client identifiers.

	-hpa