Sebastien Roy wrote:
> On Fri, 2008-07-25 at 12:25 -0400, Peter Memishian wrote:
>
>>> In another follow up, it occurs to me that the DR problem could be
>>>
>> > solved different for GLDv3 NICs. GLDv3 drivers don't process DLPI
>> > messages directly, but rely on GLDv3 to do it for them.
>>
>> There are a lot of possible solutions if you're willing to redesign the
>> framework, hence my original comment.
>>
>
> Right, and I agree with your original concern. I think the ideal
> behavior would be one that allows DR operations on a NIC while
> associated datalinks are open. IMO, GLDv3 should handle this under the
> covers while there may be references to the datalink, and the link state
> would reflect the fact that there's no underlying hardware.
>
Actually, I once worked on a project called "Alternate Pathing", that
provided "redundant data links", so that you could collect a group of
NICs together, and have them virtualize it. Much like what aggr does
today, except that instead of active-active it was active-passive. The
software worked with DR (and later RCM) to coordinate its actions, so
that if a redundant link was available, the backup link was activated
automatically.
This had the very nice feature of enabling DR to operate, with zero
impact on service delivered. (It even worked with snoop, e.g. you could
be snooping traffic arriving on one hme port, and if it was part of an
AP group, you could remove the hme port and see snoop still running.
The only delays were switch derived where the switch had to figure out
that your MAC address had moved to a different port.)
Alas, AP was killed, as at the time Solaris Networking was promising to
subsume the functionality in IPMP. However, IPMP failed to deliver
redundancy for anything other than IP datalinks.
Still, even today, if we took the position that it was not possible to
open a link that had been configured as part of a redundancy group (I
guess in aggr, though to be fair, I'm not entirely sure how aggr would
deal with this), and allowed for aggr to deal with device removal
(perhaps using RCM coordination), then we could live with the current
rules, even with reduced privilege access to data links. It would just
be necessary for admins that care about DR to configure link redundancy
in aggr. This is actually the historical AP model for high end servers,
and is better than allowing detach of active data links, IMO.
-- Garrett
> -Seb
>
>
> _______________________________________________
> nwam-discuss mailing list
> nwam-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/nwam-discuss
>
