Impossible SCSI Question

[Jason Berendsen]

I have 2 N class servers connected to an HP EX1200. One servers scsi initiator is set to 7 while the other is set to 6. HP doesn't support this configuration and Filenet (The third party software that does support this configuration) hasn't been able to give us input to get this to work. While the primary server owns the 1200 everything works fine. Then the secondary server owns the 1200 nothing works unless we change the SCSI ID of the initiator to 7.

So to the impossibility. Only the primary needs access to the 1200 until there is a hardware problem and Service Guard fails the functionality to the secondary server. For the interim I would like to see if there is a way to unconfigure the SCSI card in the secondary server and set the SCSI initiator to 7. Then when the primary server fails I would like to claim the SCSI card and at the same time claim the 1200. It's wonderfully easy with LAN but is it at all possible with SCSI?

Thanks,

Jason

[melvyn burnard]

As per my previouss response to your previous question, the answer is no.
The only supporetd way to do this is using ATS in ServiceGuard or I believe the FileNet application has a way of doing this.

[Dietmar Konermann]

Hi, Jason!

I agree with Melvyn... the configuration is not officially supported.

But the symptoms point to some trouble with the jukebox' contoller acting in multi-initiator environments. Reading through the fix list of the latest firmware there are known problems in this area. At least the fixes make me believe that the controller SHOULD be able to run in multi-initiator configs.

What you could test is powercycling the jukebox directly before the 1st request from the "ID-6" side. In this case I would recommend upgrading the firmware.

Another workaround would be using a FC-SCSI bridge. With this config only one initiator (the bridge) talks to the jukebox on a single bus.

Regards...
Dietmar.

[Jason Berendsen]

melvyn has recommended using a fibre bridge, but in this day and age of no budget it may be hard to come up with the funding. Our 3rd party software company (Filenet) is looking into why this doesn't work, but I am trying to find any type of interim solution.

I have powercycled the jukebox before the secondary server takes control to no avail. Does this rule out the firmware upgrade possibility resolving this issue?

[Dietmar Konermann]

Hmm... this sounds really weird.

Could you please try to access the picker directly via the path-through driver:

# mknod /dev/rscsi/c10t5d0 c 203 0x0a5000

(assuming 203 is the result of 'lsdev -d sctl').

# mc -p /dev/rscsi/c10t5d0 -q

# mc -p /dev/rscsi/c10t5d0 -r DS

Best regards...
Dietmar.

[Jason Berendsen]

Dietmar,

Unfortunately, because the primary is working without a problem the powers that be want to wait until the weekend when we are out of production to test the bus with the command sets you sent. Until then, what is your theory?

[Michael Lampi]

If your second server has no need to access the shared SCSI bus when the primary server is operational, then an alternative is available: a SCSI switch.

With a SCSI switch it is possible to have both servers have their initiator address set to 7, as only one server will be switched onto the peripheral SCSI bus at any given time.

When the primary server fails, the SCSI switch will disconnect the primary from the peripheral SCSI bus and connect the secondary server to the bus.

See http://www.apcon.com for details, and if you go this route tell them I sent ya!