Raoul Carag wrote:
>
> Cathy Zhou wrote:
>>>>
>>> The information was provided to me by a writer who manages hardware
>>> documentation. It's for this reason that I also maintained that
>>> separate section about replacing network cards separately (Section
>>> 4) from this section because I thought that DR is a different
>>> operation altogether. My understanding about DR is that, if it is
>>> supported, one can make hardware swaps without any downtime for the
>>> machine. So there are laptops that can do this as well? I'll look
>>> into DR more and then revise the doc.
>>>
>> Yes. Actually Joanna even tried to test DR on laptops. As it is more
>> available than DR capable servers.
>>
> I think we can make a statement that if the system supports DR, then
> what we describe about DR procedures with link names apply. The
> information that was provided by the hardware writer covers hardware
> that Sun manufactures. For non-Sun machines, including laptops, user
> should check to make sure that DR is supported in those systems.
All systems with cardbus support DR.... cardbus hotplug _is_ a form of DR.
Systems with other hotpluggable busses exist ... SATA, SAS, FCAL, and
even SCSI can be hotplugged if appropriate actions with cfgadm are taken.
Lets not forget about USB and IEEE1394 busses either.
>
> This leads me to a related question. Does swapping hardware and then
> renaming links limited only to systems that support DR? Do the
> procedures apply when you remove and replace NICs, e.g., by turning
> the system off? This might be the sense of the procedure in Section 4
> ("How to replace a network card while using link names"), which, in
> turn, was based on the current docs.
Now I'm sorry that I've missed most of this thread. But it sounds like
what you're getting at is what the old Alternate Pathing product I
worked on did... effectively creating a "meta" link-layer device built
upon different physical links. In response to DR, we used to use an RCM
script to "switch" which meta device was active if the active interface
of a metanetwork group was being removed.
Our E10K customers were very happy with this approach. (This was in
Solaris 7 and 8, before IP multipathing made its debut.)
Lets talk offline if you have any more questions about this.
-- Garrett
