[systemd-devel] RFC: Predictable Network Interface Names

Kay Sievers kay at vrfy.org
Wed Feb 13 10:45:31 PST 2013


On Wed, Feb 13, 2013 at 7:35 PM, Tomasz Torcz <tomek at pipebreaker.pl> wrote:
> On Wed, Feb 13, 2013 at 07:29:10PM +0100, Kay Sievers wrote:
>> On Wed, Feb 13, 2013 at 7:19 PM, Tomasz Torcz <tomek at pipebreaker.pl> wrote:
>> > On Tue, Feb 12, 2013 at 09:07:00AM +0100, Jan Engelhardt wrote:
>> >>
>> >> On Monday 2013-01-07 23:29, Lennart Poettering wrote:
>> >> >
>> >> >For your example the new code would pick a name of "enp0s0d0", i.e. for
>> >> >pci bus 0, slot 0, and dev_id 0.
>> >>
>> >> Is it Solaris time yet? enp0s0d0, that's just like c0t0d0s0. And
>> >
>> >   Heh, just in time when Solaris 11 deprecated such names, choosing
>> > to name interfaces as net0, net1, net2....
>>
>> The Solaris disk names are based on meaningless unpredictable
>> enumerations, they have never been topology based. They are just 3
>> levels deep, which made them only slightly better than a single
>> number, it's still the same useless enumeration with "inventing"
>> artificial numbers.
>>
>> I don't remember seeing topology based names for network interfaces on
>> Solaris, they did that?
>
>   Not really, just "driver name""detection order" - ixgb0, nge1 etc.

Ah, ok. Every driver had its own counter starting at 0, counting
upwards for every new instance? Yeah, makes the problem quite a bit
smaller than the one global ethX counter we use on Linux.

"Nice" that they now switch to the most fragile solution they can choose. :)

Kay


More information about the systemd-devel mailing list