TAPE--BEYOND Performance Factors.

In these pages, for about 18 months or so, I have explored many tape technologies. These covered the gamut from the ridiculous (QIC) to the sublime (the still unavailable Ultrium and SDLT and currently available Mammoth-2). Clearly, each drive was (or will be) most suitable for a subgroup of users or applications.

For example, a Travan5 drive with 10GB native and 20GB compressed capacity may be fine for backing up a desktop workstation with only a few gigabytes of important data, while a DLT 8000, AIT2, or Mammoth 2 drive would be a better fit in an enterprise that required lots of data backed up and didn't have a lot of time to wait. A DDS3 or 4 changer might be a good choice for backing up a small network--with one drive and up to seven tape cartridges such as DDS4 devices could, conceivably, back up as much as 140GB of uncompressed data and some unknown amount of compressed data.

The phrase "speeds and feeds" is applied to a lot of products--including tape. Megabytes per minute remains a good yardstick for comparing tape drive performance, even if it's expressed in compressed and native format numbers. Native and compressed capacities are another good yardstick for determining the suitability of one technology when compared to another.

I am still, personally, unconvinced about claimed compression ratios. There are only a few compression algorithms currently in use. Further testing may be able to determine whether an algorithm that claims an average 2:1 compression actually stores less data or writes more slowly than does a drive with a claimed 2.3:1 average ratio. Perhaps the readers who have tried both can tell me their experiences on this issue.

Last month, we looked at factors affecting performance. The servers attached to the highest performance drives, for example, may be unable to send or receive data at rates matching those of the drives.

Last month, I discussed some of the ways an IT manager can minimize the speed bumps and get the best possible transfer rates from the system being backed up. These included putting the tape drive on its own SCSI bus--not having to compete with other devices on the bus can help avoid a SCSI logjam.

Another suggestion, which I now am rethinking, is to defragment the drives being backed up. The idea of optimizing drives, on its face, sounds like a reasonable way to improve the reading performance of the drive. There may be some problems with this approach. First, it's a good practice to back up any drive that you plan to optimize before attempting optimization, in case something goes wrong with the drive. If you're backing it up so that you can optimize it, so you can back it up again, this hardly makes much sense. Any data that is written to the drive after it is optimized will contribute to fragmentation of the drive. You can't win.

Worse still, in a recent issue of CTR, it was made clear that "defragmenting" a drive may not improve performance and can, in some cases, fragment it even more. Trying to defragment a drive that has multiple partitions or a RAID array doesn't make much sense, either. When the data is striped across multiple drives, what does defragmentation do to the RAID array?

The point should be clear--selecting a drive based solely on performance may be overkill--especially if the system being backed up can't match the performance of the drive. Choosing a drive or system based on capacity might be a better criterion--at least you should be able to get all your data backed up. For mission critical backup in environments with very small or nonexistent backup windows, maximum performance may be crucial. Fast access to all your data is critical.

One factor in selecting a drive technology (make no mistake, it's a technology that you're selecting and once the choice is made, this can be the basis of a long-term marriage). The other factor is price. Until the time comes when all businesses have unlimited budgets for backup or the drives become practically free, price will continue to be a major part of most tape purchase decisions.

Some Notes On Products

In November, Benchmark Tape Systems previewed its seven tape changer, which married a Benchmark DLT-1 drive and up to seven tape cartridges to a changer manufactured by ADIC. In mid-March, Benchmark began shipping the device.

Benchmark isn't the only "inexpensive" drive manufacturer to offer its own automated backup system. At press time, Ecrix announced the AutoPak, a changer that marries one or two VXA 1 drives to SpectraLogic's TreeFrog changer. The AutoPak can be used as a desktop or rackmount unit.

The basic AutoPak, with a suggested retail price of $4,495, includes one drive and support for up to fifteen cartridges. Native storage capacity for the AutoPak is 495GB (15 X 33GB) with a claimed compressed capacity of 990GB. A second drive and a bar code reader are optional upgrades to the AutoPak. Additionally, the VXA drive is available with a single-ended SCSI-2 interface or with an LVD interface.

SpectraLogic, which manufactures the Tree Frog changer used by Ecrix in its AutoPak, also manufactures a version with a Fibre Channel interface. At the time of this writing, Ecrix did not announce or offer the fibre version of the AutoPak. In some ways, using a Fibre Channel interface for a tape drive may not make a lot of sense, if one looks entirely at specs. However, for companies that have made an early commitment to fibre and are looking more for compatibility than performance gains, a fibre version may make sense. Although Ecrix has done the integration work for its SCSI version of the AutoPak, a future fibre version may not appear to be completely Out of the question. Although the current release of the AutoPak provides higher total capacities than Benchmark's autoloader, both drives are also available in automated systems from other vendors that provide higher storage capacities.

The DLT-1 is an interesting product--it has the capacity (40GB native) of a much more expensive DLT 8000 drive, but half the transfer rate (3MB/sec for the Benchmark drive vs. the DLT 8000's 6MB/sec native). The DLT-1 can also read (but not write) a tape created on a DLT-4000 drive.

At a price less than one-third that of a DLT-8000, it may cause some users to consider their backup and performance requirements. If, for example, an organization typically performs backups of less than 80GB (and many incremental backups probably fit this parameter) and the backup is done unattended, overnight, a DLT-1 may be a good choice.

Further, with an autoloader, the choice may be even clearer. An interview with an automation vendor brought up another issue relating to the drives used in automated libraries--the cost of the drive(s) represents a large portion of the total cost of the library. Reducing the drive cost can either reduce the overall cost of the library or can make the purchase of a second or third drive, with certain backup performance improvements resulting from the use of multiple drives, an obvious choice.

The comparison between low-priced drives destined for midrange backup applications doesn't stop with Benchmark. The Ecrix VXAl drive, recently reviewed in Computer Technology Review, at under $1,000, provides the user with a similar quandary--especially where automation is concerned. The VXA1 is an 8mm drive, using media and a format that are not compatible with such other 8mm drives as those from Exabyte (the Mammoth 1 and 2) or Sony (AIT1 or AIT2). However, at under $1300 and probably less when the raw drive is married to a library or jukebox, the VXA1 provides a cost advantage and a claimed reliability advantage. The VXA1 uses a packet writing format, which Ecrix claims to make recovery of data possible, even from badly damaged tapes.

Where cost is a factor, both the DLT1 and the VXA1 drives can be important options, even if bought as standalone drives for backing small networks or large workstations. It should be noted that Quantum owns a significant portion of Benchmark and Benchmark licensed certain technologies from Quantum, so a win for Benchmark may still be something of a win for Quantum.

All this being said, there's still much interest in the as yet unreleased high-end drives. Both the Ultrium and the Super DLT (SDLT) drives are expected in the next few months or later this year. Both promise 100GB native capacity (although lower capacity media may be offered) and both have top-of-the-line performance. Further, both technologies will quickly become part of automated systems, providing high amounts of data storage per square foot of floor space (although 8mm may still beat them in the capacity/floor space spec).

The new Ultrium and SDLT bring additional benefits when compared with DLT. Both deliver higher capacities and higher performance. Both provide a technology roadmap that will extend many years into the future.

The tape cartridges will not be subject to the potential of making the cartridge unusable or damaging the drive that is legendary with DLT. Redesigned (or newly designed) latching mechanisms have overcome this problem.

Additionally, memory in cartridge will enable the drive to write catalog and other information regarding the cartridge onto the cartridge during use and before ejection. An optional reader can recover this information without loading the cartridge into the drive. This feature may be especially useful for automation--a library can be cataloged in minutes rather than hours simply by passing the reader near each cartridge.

It should be clear by now that there are many different factors to be considered when choosing a tape drive technology, whether standalone or automated. The introduction of new, low cost alternatives makes the picture somewhat more confusing. High end Ultrium and SDLT drives and the current Mammoth 2 and AIT2 drives further complicate the decision.

With a careful consideration of many factors: capacity, performance (and whether or not your system can provide data to the drive fast enough), cost, anticipated future requirements, and platform longevity (do you want to upgrade to new drives in two years? Will new drives be available?), support, and service are all factors to consider. Careful consideration of all these factors will help to make the right choice possible, if not still pretty difficult.