> Sounds insane. Maybe SAN allows this, but it sounds foolish. Too many
> ways to cause problems.
No kidding :-).
> > This is why I stand by what I said- the driver (or some midlayer)
> > has got to either assure a address constant to the app or fs layer
> > while the system is running or send back an I/O error if it can't.
>
> But will the disc *I* plug in have the same host,bus,target,lun?
> And if not, how on earth do I find out what the unique identifier is?
> I've just walked downstairs and plugged in 5 new discs, meantime
> someone in another building has plugged in another 5 new discs.
>
> Which are mine?
The ones with the WWNs you've written down :-).. Seagate && other folks should
be putting WWNs on the disks themselves- I haven't seen them there, but maybe
they're starting to do this- some HBA vendors do this for their cards' WWNs.
Look- when you plug FC-AL disks into a loop, they don't necessarily go to
fixed addresses. They tend to go in the same spot in a chain- they try to
renegotiate their old loop address, and since loop position allows you to pick
the addresses not taken, you can somewhat predict how things are in a local
loop. But you can't entirely predict or set address (at least from the point
of view of planning for a large number of configurations, assume you can't).
A fabric is a harder problem since this is lots of loops connected together
and has an expanded multilevel addressing schema (which does have some
physical topology correspondence).
What you do, or will do, in practical terms, is you've plugged a box of disks
into a switch. You use management software or a telnet to the switch switch to
confirm that they're there and being seen.
>From this you can get the WWNs so that you can, when you rescan from the
system you're running on so that they're seen from your system, get their
local 'name' (whether it's /dev/sd? or the fancy new /dev/wwn/foo...).
This is what you can use to do the necessary volume import where you get them
labelled from which point on you don't care particularily about any
addressing. Or if it's not a device that you can label (like a tape drive),
you retain the WWN and use that.
A local loop, w/o the fabric, is much like this except you don't have the
confirming step of checking with a switch and say- let's see, what just got
added to port 7? Instead, like a local SCSI bus, you have more control over
what you do/don't plug in so you can pretty much do an A/B comparison to see
what just arrived.
This is all a big pain in the buttinsky, I agree. The alternatives aren't
pretty either- some are like what NT's Disk Administrator does- it just grabs
*everything* it can see and says, "Can I write a signature? Please? It won't
hurt anyone! Trust me!".
-matt
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Apr 30 2000 - 21:00:17 EST