Still using disk pool with VTL?

[avinim8]

Hi,
I just need to verify a thing.
If I use a VTL in my environment, I still need to use the disk pool right?
It is the same concept as before (backup to disk and migrate date to VTL).

Is there a way to perform multiplexing in TSM - so that SEVERAL nodes will backup directly to tape without staging first with disks?

Thanks,

Avi

 

[toxy13]

You can realize it with a seperate "copygroup" and there you define the "destination" parameter to you VTL pool.

 

[cjhood]

Avi,

You don't necessarily have to have a diskpool in front of a VTL. If you've defined enough virtual drives you can have all your backups going directly to the VTL. It ends up being sort of like a diskpool, only now each backup session needs one (or more) virtual drives to write to instead of all of them streaming at the same time to disk.

If you drop the diskpool, you will need to check what the max ingest rate is for your VTL, and how many clients will be backing up to it at the same time. If you throw too many clients at some types of VTL they will throttle their speeds and blow your backup times out.

You can certainly backup nodes directly to tape, however it's an all or nothing setting in the copygroup. Each client will need a drive to back up to, and I believe that only one session can use a drive at a time.

-Chris

 

[avinim8]

Is it recommended to back up directly to tape (one tape per one client)?
It seems that there should be no problem with this.

Anyone?
pros and cons?

 

[cjhood]

Depends on the type of client and if it's VTL or physical tape. I find clients with small numbers of large files (exchange, databases etc) work well going directly to physical tape, as they stream better. Clients with large numbers of small files work better going to a diskpool (or to a VTL).

-Chris

 

[lapfrank]

Hi,
I just need to verify a thing.
If I use a VTL in my environment, I still need to use the disk pool right?
It is the same concept as before (backup to disk and migrate date to VTL).

Is there a way to perform multiplexing in TSM - so that SEVERAL nodes will backup directly to tape without staging first with disks?

Thanks,

Avi

After lots of testing, I found that keeping most backups to a diskpool was better than sending everything to tape. First, TSM doesn't fare well when managing a lot of mount points (diskpools do not create mount points) for tapes, so creating 120-150 virtual drives was really not a good idea. I've set up 40 virtual drives for the VTL and only send the biggest backups directly to virtual tapes (and the ones that use lan-free):
- domino / notes data
- exchange data
- oracle data

for the rest, I still use the diskpool with a high threshold for migration that is set pretty low (around 50%).

It's been working well.

 

[GregE]

When you say diskpool are you using random disk or sequential? (file devclass)

 

[lapfrank]

Random disks. I have less than 10 storage pools, so I do not need FILE class (which has mount points, and therefore causes unneeded loads on the TSM Server). I can arrange my disk space with small 10 to 50 gigs raw volumes between my disk storage pools.

Not sure the FILE device class would have a lot of advantages over the virtual tapes, as it also uses a mount point and is ultimately disk. The biggest one would be the limited ingest throughput of your VTL..