Re: Specifying names for tap interfaces

[Mouse <mouse%Rodents-Montreal.ORG@localhost>]

>> Perhaps I did conceptually misspeak, you can call it a UID if you
>> want; the concept is the same, an identifier that is entirely
>> unlikely to be represented in the same machine, [...]
> Sure it is, [...]

For port-xen's purposes, perhaps, provided you're talking about real
hardware, because xen is rather restricted as to what hardware it runs
on.  For tech-net's purposes, no; MAC addresses are required to be
unique per-host, not per-interface, and at least one of the more
popular second-class ports - sparc - normally uses the same MAC on all
the interfaces on a given box.

> but what about virtual interfaces like tap(4)?  We are not talking
> about hardware NIC, but virtual ones.

That too.  With proper administration, this is not an issue, but I
suspect few admins want to go to the trouble of (a) actually keeping
track of their local MAC use and (b) configuring locally-administered
MACs on everything relevant.

>> Virtualization-platform-providers are given the same individual MAC
>> prefixes as anyone else building network interfaces.

Well, those that have enough money to spring for one.  OUIs ain't cheap
enough to be trivial for a volunteer open-source project (the figure I
remember is $10,000, but that memory is pretty fuzzy).

> Which lets 3 bytes of randmoness for the suffix.  Not a whole lot if
> you want to avoid collisions given the birthday paradox.

Yeah, you can expect a collision somewhere around 4K values, and if you
want the chance to be decently small you have to cut off well below
that.

But, on the other hand, the people running large farms of VMs are more
likely to be willing to go to the trouble to administer
locally-administered MAC space, which gives them 46 randomizable bits.

/~\ The ASCII                             Mouse
\ / Ribbon Campaign
 X  Against HTML                mouse%rodents-montreal.org@localhost
/ \ Email!           7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B