Agreed, great thread

We currently have a SUN VTL that has NDMP functionality. i.e. the physical drives are zoned into the vtl only. Add the drives are configured as NDMP drives.

The idea was to utilise SLP’s for the primary write to the vtl and duplicate off to physical with the VTL acting as the data mover.

In theory all this was great but it turns out that the VTL NDMP component is incredibly flakey and we have now had to resort back to traditional methods.

This is a great analysis... However, one thing a VTL is good for is handling slow clients.  For me, backing up about a hundred clients to 12 virtual tape drives significantly shortened my backup window compared to backing up to 4 LTO3 drives (I have quite a few slow clients to contend with so backing up straight to the LTO3 drives was never an option anyway).

I didn’t like how NetApp implemented their direct to tape feature so I use Vault (soon to be Storage Lifecycle Policies) to duplicate images from the VTL to physical tape.  I set aside a 5th LTO3 drive for restores.  I get great throughput with the Vault jobs and it’s pretty much trouble free.

Neil


On 10/26/09 11:42 AM, "Ed Wilts" <ewilts AT ewilts DOT org> wrote:

On Mon, Oct 26, 2009 at 1:00 PM, Heathe Kyle Yeakley <hkyeakley AT gmail DOT com> wrote:

Here's my current layout:
Hardware: Spectra Logic T380 with 12 IBM LTO4 tape drives.
                    * I'm told the theoretical bandwidth of an LTO4 tape
drive is approximately 120 MB/s. If I have 12 drives, I'm assuming I can
say that my library should theoretically be able to handle data at (12 x
120 MB/s) 1,440 MB/s.

Not quite true.  You can theoretically deliver UNCOMPRESSED data to tape at 1.4GB/sec.  However, if you are getting 2:1 compression, you can deliver twice that rate - 2.88GB/sec.

If you're getting 3:1 compression, you can deliver 360MB/sec per tape drive.  That's 1 dedicated 4Gbps fiber channel port per drive.  For 12 drives, you can possible write at 4.3GB/sec depending on your compression ratio.
 

NetBackup: 1 Master (linux), 2 Media (linux), and 3 San Media Servers (Tru64).

I suspect you don't have enough media servers to be able to deliver the data rate to keep the tape drives busy.
 

* I have a "library" that can receive data much faster than my network can deliver it.

Welcome to the club :-(
 
The first misunderstanding is that disk is faster than tape.  In general, it isn't.   You'd be hard pressed to find a disk subsystem that your management is willing to pay for that can actually write at 4.3GB/sec.  Management typically thinks that just because they're backups, you can use cheap (SATA) disks without really acknowledging that backups can have the highest I/O workloads of any application that the company runs.

Assuming your master server does no tape I/O, you have 5 media servers to put data to tape.  That's 2-3 tape drives per media server.  You will need at an absolute minimum a pair of HBAs dedicated to tape work - and pray that NetBackup actually will give you 1 tape drive per HBA, which it doesn't even try to do - and you'll need to receive at least 240MB/sec of network traffic - i.e. 3 GigE connections for those non-SAN media servers. 

Performance management is tough and unless you can clearly identify the bottlenecks you have now, buying hardware is NOT the right answer.  You may very likely be throwing money at the wrong problem.  If, for example, you don't have enough media servers, you'll have wasted the money on the VTLs.

Without de-dupe and assuming 2:1 compression, the VTL 1400 can injest 8.2TB per hour.  That's 2.3GB/sec and is actually *LESS* than what your 12 LTO-4 drives are capable of with 2:1 compression (2.8GB/sec).  With de-dupe running, you'll get about half that performance out of the VTL (4.3TB/hr)

On top of all that, the direct tape creation speed is rated at 3.0TB/hr.  That's 0.8GB/sec and now you're significantly less than what your existing tape drives are capable of.

"why did I buy a  VTL? I haven't gained anything from what I see."

It depends on the problem you're trying to solve.

Assumptions: One of the principle reasons anyone deploys a VTL is because:
      * Disk is faster than tape at the expense of  disk not being removeable and having a lower Mean Time Between Failures than tape.
      * Backup windows are shrinking. A VTL allows you to create several "virtual drives" that allow you to write more backups concurrently, thus shrinking your backup window.

Both assumptions are actually false.  Disk is not faster than tape (see the math above) and although your backup windows are shrinking, a VTL may not allow you to write more backups concurrently. 

VTLs may solve problems, but backup performance in large environments is not one of them.  Restore performance is one area where they do help.

   .../Ed

Ed Wilts, RHCE, BCFP, BCSD, SCSP, SCSE
ewilts AT ewilts DOT org

_______________________________________________
Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu

Notice to recipient:
The information in this internet e-mail and any attachments is confidential and may be privileged. It is intended solely for the addressee. If you are not the intended addressee please notify the sender immediately by telephone. If you are not the intended recipient, any disclosure, copying, distribution or any action taken or omitted to be taken in reliance on it, is prohibited and may be unlawful.

When addressed to external clients any opinions or advice contained in this internet e-mail are subject to the terms and conditions expressed in any applicable governing terms of business or client engagement letter issued by the pertinent Bank of America group entity.

If this email originates from the U.K. please note that Bank of America, N.A., London Branch, Banc of America Securities Limited and MBNA Europe Bank Limited are authorised and regulated by the Financial Services Authority.  For all U.K. corporate disclosures, please refer to www.bankofamerica.com/ukcompanies