James Carlson wrote:
> Francesco DiMambro writes:
>
>> The discussion took a lot of twists and turns and lasted longer
>> than I had expected. Along the way I never got the main question
>> I was looking to be answered; which was whether I could get a
>> contract drawn up allowing me to more officially use MDT.
>> You did mention someone I could contact about this, Markus
>> Flierl, would his e-mail be '[EMAIL PROTECTED]'?
>>
>
> Yes, that'd be the guy.
>
> (My guess would be: "the team that works on that code has no interest
> in maintaining it, and is even in the process of ripping it out, so
> unless the contracts are pretty clearly delineated in a way that it
> doesn't affect Nevada/OpenSolaris in any way, they'd probably be met
> with some resistance." But I'm not on that team, so it's just a
> guess.)
>
>
Furthermore, IMO, if we want/need MDT (or a workalike), then I think the
right place for any contracts to be made would be at the GLDv3 layer.
Of course, the GLDv3 itself is still not public, and lacks any kind of
MDT support right now, in favor of LSO. But that could change if there
was a sufficiently strong case made for it, I think. I hope that part
of any such change would be fixing the ugly portions of MDT that have
their fingers spread throughout the networking stack, so that non MDT
drivers don't pay a penalty for the enablement of MDT by a few select
drivers.)
FWIW, I abhor the idea of creating additional, or extending via
contract, networking interfaces which require the driver to implement,
or even be aware of, the DLPI. (It is also true that you wind up
leaving some performance on the floor when you go the route of pure DLPI
-- there are specific optimizations only available to native GLDv3
drivers.) But like Jim, I'm not on that team. And unlike Jim, I don't
even report into the SNT organization anymore, so my feedback probably
carries even less weight.
-- Garrett
_______________________________________________
networking-discuss mailing list
[email protected]
