Re: [9fans] unique MAC address allocation?

[Michael Zappe <zapman@zappe.us>]

Derek Fawcus wrote:

>On Wed, Mar 30, 2005 at 11:00:08AM -0700, Michael Zappe wrote:
>
>
>>As to question 4, nope.  The OS has no good way of determining a new MAC
>>address to override to (The OS doesn't have an OUI, and then how do you
>>pick the serial number after that??),
>>
>>
>
>Pick a random number,  set the local allocation bit (second bit on wire)
>and then test the address to see if anyone responds to it.
>
>Didn't DECnet do something like that?
>
>
>
Yeah, but you're not guaranteed a reply to the packet.  It's a distinct
possibility that if you pick an address and a protocol, the other
machines stack may disregard it, i.e. blackholing pings, etc.  There's
no good way to detect the collision.  Theoretically you could use LLC to
check, but I wouldn't count on that being implemented correctly, if at
all... :-)  If you're using locally administered addresses, you can
assume that valid NICs aren't going to have this address, and the chance
of a collision is low. (N in 16 million for N machines with the same
OUI.)  So I guess that would work! :-)

DECnet actually had a deterministic algorithm for assigning MAC
addresses.  You had an area number that was 6 bits, and a 10 bit node
number.  Then you appended this to AA:00:04:00 to get the MAC address.
Multiple interfaces on the same machine all used the same MAC.

>DF
>
>
>
>>and it's not considered a real
>>problem because of the uniqueness of MACs.  Not to mention you don't
>>even want to think about how a switch could get confused by the whole
>>situation... :-)
>>
>>    Mike
>>
>>