Multiple TSM Servers with single IBM 3584 (TS3500)

[jharris]

Guys,

We've been running an enterprise TSM environment with 7 or so TSM servers communicating to a TS3500.

A single TSM Library Manager instance was created some time ago and manages all of the mount requests from the other TSM server library clients.
Our TS3500 is ALMS enabled and also has attached a couple of C06 (IBM 3593) controllers for IBM Z10 Mainframe to share the library with seperate dedicated tape drives.

We were speaking with an IBM engineer this week who suggested we should ditch the TSM Library Manager and use the ALMS feature built into the TS3500 for library management. Is anyone doing this and how does this effectively change our environment, keeping in mind our TSM Library Manager has a library definition of type SCSI and the TSM library clients are type SERVER.

Cheers.

 

[TonyB]

Like yourself, I'm currently running a ts3500 w/ ALMS, multiple TSM instances, a mainframe tape controller (C06) with a TSM library manager (several, in fact).

I don't understand how this IBM engineer's idea can work.

Presumably (s)he's referring to configuring drives as shared between multiple logical libraries, with each "media server" (if I may use that term in a TSM forum) having its own logical library.

How the &#*$ is that supposed to work? There's nothing serialising access to the drives. Where is the piece of software that decides that F01D01 is currently assigned to the C06, and doesn't even offer it to a TSM instance as a viable mountpoint? I reckon you'd be looking down the barrel of a heap of failed tape mounts.

Hey, maybe I'm wrong - but the last time I looked at ALMS you could share drives around, but only by ensuring that apps upstream of the shared pool don't try and use a drive at the same time (i.e. by disabling drives selectively).

 

[Nicke]

In "jharris" case you actually need a ALMS partitioned library with differ between the Mainframe environment and Open.

Also it works nice to utilize both ALMS and TSM Shared Library with Library Manager(s) and Library Clients(s) within an ALMS partition.
But as previously noted I don't see any working solution with Shared drives in ALMS, use dedicated drives instead.

Kind Regards,
Nicke

 

[chad_small]

The IBM engineer doesn't know TSM too well. If you ditch the library manager you'd have to partition the library and assign each instance specific drives since even ACSLS can't manage multiple TSM instances. You have to have the library manager.

 

[jharris]

Putting the mainframe aside ... I think what was being implied was that a virtual library is defined on the TS3500 with ALMS enabled for each TSM server instance. Then the common 9 physical tape drives we have for TSM usage, get shared between each virtual library.
The downside is that we lose a common scratch pool, excess scratch tapes would need to be checked-out (not removed) from one virtual library, then checked into another virtual library if needed.
But ... we could do away with common TSM Library manager instance we have running.
I'm going to do a little research into this.

 

[TonyB]

I suspect that even with the mainframe out of the picture you might see serialisation problems...

Lets posit 2x TSM instances (each a library manager) with 2x logical libraries, and 1 drive shared between the logical libraries.

At t=0 instance 1 mounts vol1 in the drive
At t=1 instance 2 attempts to mount vol2 in the drive and gets back an error (drive is already in use)

Since the two instances think they each have a unique tape drive (the single shared one) each will attempt to mount tapes in it whenever they want. Some of the time this will work...up until the point that they both want the same physical resource at the same time. Maybe (just maybe) ALMS will block the incoming mount request and queue it - in which case you'll just see a very long mountwait. Even if it does do this - it may not be desirable. Say we have 10 drives, one is busy (instance 1, which is running a 10 hour backup job)...instance 2 decides it wants a drive for a 1minute backup job - murphy's law states that it will want the one that instance 1 is using - it then queues a 1 minute job for 10 hours...while there are 9 drives free.

 

[chad_small]

I can usually run a TSM library manager instance with a 2GB DB and 1GB log and that's big. I had a 7 instance environment at my old employer too. I wouldn't even consider using ALMs instead. The loss of funtionality would severely impact performance and increase my headaches.