Feasibility of using Y-SCSI cable to connect direct library picker

[Steven Sim Kok Leong]

Hi,

I have an MC/ServiceGuard cluster of two omniback cell servers which are also acting as media agents, with an Omniback package configured.

During peace-times, the primary node has a SCSI cable connecting it's SCSI controller directly to the library picker of a tape library. When a failover occurs, the SCSI cable has to be detached MANUALLY from the SCSI controller of the primary node and attached to the SCSI controller of the adoptive node.

My question is, since only one node is active at any one time, thus only one node will be sending SCSI instructions to the library picker at any one time. By ensuring that only one node sends the SCSI instructions, I felt that it is alright to have a Y-SCSI cable (I do not know the exact name for it) with the top part of the Y connected to both the 2 nodes' SCSI controllers and the other end connected to the library picker.

HP told me that they are not going to support such a configuration. However, I am thinking of hooking up the Y-SCSI cable myself, otherwise, there is really no point for me to ServiceGuard cluster two Omniback cell servers when the failover still has to be MANUAL at the picker portion. It defeats the purpose of having ServiceGuard in the very first place.

Are there any issues that I might have overlooked or any concerns I need to take note of with regards to hooking up the Y-SCSI cable?

Btw, we do not intend to purchase ATS (Advanced Tape Services) or migrate to Switch Fabric, so SCSI connections are still here to stay in my environment and library picker limitations apply.

TIA for any help. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Steven Sim Kok Leong]

Hi,

Any Gurus out there willing to give me a hand at this? :)

TIA for any help. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Patrick Wallek]

What about a SCSI switch?

Here are a couple of links to scsi switches at Black Box.

http://catalog.blackbox.com/BlackBox/templates/blackbox/itemgroup1991guest.asp?param=270&ig_id=1991&title=6x4+Electronic+SCSI+Switch&related=

http://catalog.blackbox.com/BlackBox/templates/blackbox/itemgroup247guest.asp?param=270&ig_id=247&title=Reversible+SCSI+Switch&related=

[Steven Sim Kok Leong]

Hi Patrick,

Thanks for your answers.

How does the SCSI switch work? Does it require any manual intevention during the "switching"? If it actually allows for an automatic disabling of the SCSI cable connected to the primary node and enabling of the SCSI cable connected to the adoptive node, will HP support it?

TIA for any help. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Patrick Wallek]

I don't know exactly how the automatic switches work. We used a manual switch at one time to switch a tape library between two Sun servers. Never had any problems with it, other than remembering to switch it at the appropriate time.

I would guess that the automatic switches would grant access to whoever asks for it first, and deny the other machine then. At least I would hope. I haven't ever used one of the automatic ones, it was just an idea I had.

[Steven Sim Kok Leong]

Hi Patrick,

What on Earth did you do to your posting? :) Did you add any malicious html tags to your posting or what? :) I am seeing two columns of the thread. In addition, for reasons I am still unclear, I am unable to assign points to any of your postings.

Guess I will just have to leave a note with the Forum administrator or hope that Laurent or one of the Forum administrators will just pass by and correct this thread.

Thanks for your answers. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Patrick Wallek]

Steven,

I am not really sure what happened to my posts. I am seeing the same thing as you. The only thing I did in my original post was try to attach two inks to web pages of the Black Box catalog. The web pages are information on their SCSI switches.

I wish I knew what went wrong there.

Sorry about that.

[Steven Sim Kok Leong]

Hi,

I am only able to assign points to your second post onwards, not to your first post.

I believe the javascript link caused the formatting to go disarray within the tags.

Vignette Storyserver bug perhaps?

Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Carsten Krege]

The offiicial HP standpoint that this configuration would be under no circumstances supported. The official solution to share tapes in a SG cluster is the Advanced Tape Sharing feature introduced in SG 11.05 and with some kernel patches for the stape driver.

The hardware required involves a fibre channel (FC) connection from each node to a FC SCSI MUX that allows a SCSI RESERVE flag to be set and thus allows an exclusive access to the tape device. The tape device is connected to the FC SCSI MUX.

For details see he "Using Advanced Tape Services" manual at
http://docs.hp.com/hpux/pdf/B3936-90032.pdf . This also shows the possible h/w configurations. Note that not all sort of tape libraries are supported (most of the big HP libraries are).

You see that the main feature is the exclusive access to the device. This would not be guaranteed in your setup (e.g. if Omniback runs on one node and tar on the other) and expose your system to data corruption problems.

Carsten

[Steven Sim Kok Leong]

Hi Carsten,

Thanks for your reply.

However, I feel that the concerns that you have raised only apply to the sharing of tape drives.

Why would sharing the library picker lead to data corruption since the library picker only functions in picking the tape from slot to drive and vice versa? If the library picker fails to pick the tape, either I get a backup done or none at all. Atomicity is maintained.

In any case, I am certain that nothing is going to use the library picker from the adoptive node at all during peace-times.

Given that I have satisfy these conditions of mutually-exclusive access, are there any technical issues that I may have overlooked or which I need to consider for the implementation of this Y-SCSI cable?

TIA for any help. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Carsten Krege]

You are right: I read that you want to share the picker device only and forgot later when I wrote my reply. :) Of course there would be no direct (see below) data corruption problems if the tapes are not shared as I described initially.

However. you might consider this scenario:

1) nodeA loads tape1 into drive to start backup
2) nodeB unloads tape1 and loads tape2
3) nodeA starts the backup and corrupts the data on tape2

Ok, I admit this might be far fetched. But the ATS feature (plus the hardware setup using a FC SCSI MUX that supports SCSI_RESERVE flags to be set) also guarantees exclusive access to the picker device and therefore prevents this, too. :) (Here we are again.. ).

Generally spoken (and absolutely independent from your specific example and question), I wouldn't expect problems using a SCSI Y-cable if it is taken care that the SCSI termination is ok on the shared SCSI bus it is used on and if there are no duplicate SCSI ids configured (default on all SCSI host adapaters is SCSI id "7").

Carsten

[Dave Wherry · Accepted solution]

Thinking about the Y cable solution, you then have two controllers on the same bus. Could that create some sort of conflict? I'm not sure. I'm just throwing out the question.
In our shop we use the FC/SCSI bridge solution and it works very well. Host A has a FC card that connects to the bridge. Host B is the same. The bridge has 4 SCSI ports. Port 0 goes to the picker on my first library and is daisy chained to the first DLT8000 drive. Port 1 from the bridge goes to the second DLT8000 in library 1. Port 2 goes to the picker on library 2 and is daisy chained to the first DLT8000 drive of that library and port 3 goes to the second drive.
Each host sees both libraries, both pickers and all 4 drives.
With the SCSI Reserver/Release on the drives we have no conflicts and it works great. Using the FC we pump a lot of data to those drives and performance is also very good.
If I understand correctly your second host is idle waiting for failover. With this bridge configuration and media agents on that host you can use it during "peace time" for backups and get some utilization out of it.

[melvyn burnard]

As per Carstens's response, the only suuported way to do this, even for the picker, is either the ATS software supplied with ServiceGuard, or if you have a SAn with Fabric login,.
Unfortunately, the second method is outside my area of expertise, so I cannot add much more than that.