Techworld talked to ADIC's Steve Mackey, director of product marketing, EMEA, about new high capacity LTO tapes and about virtual libraries. Speed-matching and bar codes figured prominently in the discussion

LTO3 tapes have a 400GB native capacity and a 80MB/sec I/O rate; a huge capacity and a fast transfer speed. Mackey thinks LTO3 has a peculiarity to do with speed-matching that may prevent its potential performance being realised.

A drive has its rated maximum I/O speed. If a connected server can't stream data to it it that fast then it slows down; speed-matches. However, there is a limit as to how slow it can go. If the server sends data slower than the limit then the drive stops until there is enough data to start it up again. This will have quite an effect on backup time, lengthening it significantly if it happens often during a backup session.

Mackey mentions a research report that found no single customer is using LTO2 to its maximum performance. "The issue is getting data off the server disk onto the LTO3 drive. LTO2 drives can speed match down to 19MB/sec. Below that it starts and stops."

So if customers are not using LTO2 drives to their maximum what chance is there that they will get the best advantage from LTO3? Mackey points out, "It's going to be interesting - LTO3, (but) how useful is it going to be? LTO3 runs at 80MB/sec and will digital speed-match down to 40MB/sec." At which point, if the server can't keep up - it stops. "There is a common misconception that to speed backup you need a faster drive; the faster the drive I/O, the faster the backup. Not necessarily. You could end up making it worse by putting in faster technology which starts and stops."

Cutting the backup window
Everyone is generally interested in reducing their backup time. Mackey suggests that the best way of doing this is to add disk into an existing backup process. Disk unreliability is countered by using RAID. A good virtual tape library (VTL), such as PathLight VX, uses disk in this way and then transfers the backed up data to tape in such a way as to stream the tapes at full speed and fill them up.

Mackey asserts that "PathLight can at least double backup performance. Restore is even better." Part of the speed increase is down to the way the disks are tuned. Mackey says, "Disks generally are tuned for random access and can, in extreme circumstances, be slower than tape." With PathLight, files are written contiguously so that they can be written fast and streamed off quickly with reduced disk head movements.

Another VTL advantage is that writing EOF (end of file) marks can be done instantly. When backing up to tape the backup application waits for the tape to write an EOF mark. This can take a long time. Good VTL software writes an EOF instantly on disk and frees up the backup application.

Increasing backup speed is not just a case of getting either a faster drive or adding in a disk caching stage. You need to add these things in a way that seamlessly works with the existing backup application and process and uses the disk and tape hardware to best effect.

Use a VTL
He says that D2D backup is best done with a virtual library rather than by adding a drive array to, for example, Tivoli: "You have then to manage the disk manually; create partitions to backup to; resize them as data grows; allocate them to backup sets, etc." In a VTL this stuff is done by the software.

The VTL produces tapes that are bar-coded correctly and so can be read by the backup application. They can be moved to another location for disaster recovery. Mackey says that with PathLight, "Almost uniquely amongst VTLs, the production of tapes is integrated into the system. The system itself also manages which data is held on disk and which on tape." User-settable policies drive this and the actual software used is ADIC's StorNext information lifecycle management product.

How do you size a VTL?
With a tape library the rule of thumb is that you need a capacity twenty times your dataset size. If you have 1TB of online data to be backed up to a library then that library should have a tape capacity of 20TB. A VTL can have varying proportions of disk and tape. A disk-only VTL sized for a 1TB data set would have 20TB of disk - clearly a very high expense and most probably not worth it.

Mackey says that the way to approach this is to understand your restore service objectives. Some things may have a restore objective of minutes, others hours and still others days. In general you would add up your fast restore dataset size and use that as a guide to the disk component size in your VTL.

Mackey says, "You may need three day's worth of file and print data on disk but 30 days worth of Oracle database. Then it migrates to tape. A lower acquisition cost means less data on disk, but it also means restricted fast restore opportunities."

He points out that a good VTL can reduce the number of backups you need. For example, with Veritas you may need an additional backup cycle to produce an archive vault backup copy. With PathLight such a copy can be produced within and by the library, saving a Veritas or Legato backup run and the associated server CPU cycles.

VTL future
Mackey believes that VTLs will meet a demand for data protection that lies outside the traditional tape automation market. He also believes that the use of VTLs will move down from the enterprise level; "We currently bottom out at 37TB. That's too big (for medium and smaller enterprises) and there will be more entrants in that market. ADIC is currently in the enterprise (VTL Market). We have an automation range that is in the SME/mid-range market. We have significant automation announcements coming this year. We see it as a high-growth area and one were going to focus on significantly."

Hint, hint. We might, therefore, hope to see a 'PathLight Lite.'