Over the past several years, system administrators have changed the way they back up NT, NetWare, and Unix servers. Instead of backing up individual servers to direct-attached tape drives, they are backing up the whole network over the LAN to central robotic tape libraries. This trend raises new issues when evaluating tape drive technologies for automated environments.
While both helical scan and linear drives have made significant advances over the past decade, helical has the edge in terms of linear bit density, track density, overall areal bit density, and data rate per head. However, no matter how interesting the engineering details may be, they are largely irrelevant to the users of tape backup. The truth is helical and linear drives have evolved to have nearly the same specifications and road maps. What really matters now is how the application of each technology affects system backup.
When automating the network backup process with tape libraries, the capacity and performance specifications of individual drives are not nearly as important as the capacity and the performance of the library system.
Two factors determine tape library capacity: tape cartridge capacity and the number of tape cartridges. Leading helical cartridges are much smaller than leading half-inch linear technologies, which means helical-based libraries can house about twice the number of drives as linear libraries. Similarly, since performance is determined by the data rate of each drive and the number of drives in the configuration, helical libraries can provide higher performance through parallel drive operation.
And that`s not all. More drives in a library means higher availability, and more cartridges means improved media redundancy through mirroring techniques. In a centralized backup/restore process, availability and reliability are key. In the case of media, it is essential since backups usually fail due to problems with the media.
When it comes to tape wear, helical has a number of advantages over linear. Linear drives record in a serpentine fashion, which means the tape is run back and forth from end to end until all tracks are recorded. On a DLT7000 drive, for example, this requires 52 back-and-forth passes to record the full tape.
Helical-scan drives, in contrast, record data in a single pass from one end of the tape to the other, which means much less tape wear per backup operation.
Moreover, helical drives are designed so that the tape rides on a cushion of air as it travels over the rotating scanner. This cushion is designed to provide precise head-to-tape contact and to protect the surface of the tape from wear. When combined with tape-tension control mechanisms, helical-scan drives also offer longer head life than linear drives.
And, finally, the rotating head structure of helical makes it simple to include an automatic head cleaner in the drives. This feature prevents debris from contaminating the heads and the media, resulting in error-free operation–something no linear drive offers.
The helical advantage
There are three additional factors that affect backup performance in real systems: data rate, start-stop performance, and search speed.
Tape drives always specify data rate as the maximum rate at which data is recorded to the tape while the tape is streaming. However, today`s systems typically can`t supply data to the drive fast enough to ever achieve these rates. When a system can`t handle the drive at the specified data rate, the drive is forced to operate in start-stop mode, which results in a phenomenon known as “back-hitching.” Data is recorded to the tape until the drive`s caching data buffer empties, at which time the tape stops. The drive then repositions the tape to the precise location of the last recorded bit of data. This process can reduce the overall backup performance of the system. Just how much depends on how long it takes the drive to stop and reposition the tape.
Just as a car traveling at 10 mph can stop faster than one going 100 mph, helical-scan drives typically perform back-hitch operations five times faster than linear drives because they operate at a slower tape speed. (In helical drives, heads are mounted on a rotating drum, which scans slow-moving tape to record data tracks. In contrast, in linear technologies, the heads are mounted on stationary blocks, which are moved rapidly across the tape.)
If a helical drive`s data buffer is properly sized, the start-stop operation imposes no penalty on system backup performance. In effect, the performance of the tape drive scales to the performance of the system. In contrast, the comparatively long back-hitch operations of linear drives create a bottleneck, which ultimately bogs down system performance.
Another advantage is helical`s ability to search for data at high speeds. The head samples portions of the track to identify data blocks and files as they pass below. This enables helical drives to accelerate the tape up to 200 times faster than read/write speeds while searching. Once the data is found, the tape stops and rapidly repositions itself at the beginning of the data.
The nature of linear recording physics, on the other hand, makes this feat impractical, so linear drives are limited to searching at speeds that are about the same as read/write speeds. The overall result is that helical drives can locate data two to four times faster than linear drives–a big advantage in individual file restores.
The movement to centralized network backup requires tape libraries to offer high capacity and storage density, uncompromised system performance, and availability. The superior drive and cartridge density combined with high reliability makes helical scan the technology of choice for automated network backup.
Steve Georgis is director of technology and business development at Exabyte Corp., in Boulder, CO.
Linear technology holds important advantages for backup storage
The advantages of linear tape data storage systems can be summed up in one word: simplicity.
Linear tape can provide superior reliability, scalability, robustness, and affordability–as well as performance, compared to helical-scan systems. This goes back to the origins of the technologies. Linear tape systems, such as Travan NS, are specifically designed for data storage applications.
Helical-scan technologies, such as DDS and AIT, have their origins in video and audio recording, and required extensive redesign to accommodate data storage needs. The design of linear tape data cartridges and drives is inherently simple–in terms of drive mechanism, tape path, head design, and tape formulation. Helical-scan technology is comparatively complex.
These differences have a tremendous effect on reliability–a key factor in data storage systems as the growing use of mission-critical data on distributed systems makes bullet-proof reliability of backup and restore systems essential. Linear tape drives incorporate fewer moving parts, fewer motors, and use non-rotating heads. The result: fewer potential points of failure.
Tape paths are another significant difference between the two technologies. In linear systems, the tape never physically leaves the cartridge and therefore is exposed to fewer environmental contaminants. Helical drives, on the other hand, incorporate more tape path guides–all of which must all be accurately aligned, making it much more challenging to achieve high reliability at low cost.
The linear Travan NS tape drive has an important inherent design advantage. The head-tape interface functions as a self-cleaning mechanism to keep contaminants out of the all-important head gap, where they can produce data errors or drive failure. The head design reduces the spacing between the tape and leading edge of the head to the sub-micro-inch range. It literally wipes all but the tiniest contaminants off the tape at the leading edge, keeping them from passing into the head gap region. In addition, as the tape passes over the head at speed, it creates a negative (or below-ambient) pressure, pulling the tape down to the head surface for better magnetic recording performance.
In contrast, helical scan heads are much more susceptible to debris passing through the head gap region and to head gap clogging from either atmospheric debris or material worn off of the media surface.
Adverse conditions are not unusual in today`s networked environment, where servers and backup tape drives are located on factory shop floors, fast-food kitchens, and home offices. The days of climate-controlled greenhouses for mainframes are becoming a smaller proportion of the computing landscape as distributed computing spreads servers to nontraditional locations, where they face a more hostile physical environment. The ability of a data backup system to function in less-than-perfect environments is becoming increasingly important.
Reliability is one, albeit important, factor in determining the suitability of a data storage system. Scalability, performance, and affordability are also important factors. Storage capacity needs are growing rapidly, and it is vital that storage systems keep pace with the capacity growth of hard drives. Linear formats have a significant edge in their ability to migrate to higher capacities in future generations.
Linear formats tend to be more conservative for a given capacity in the areal density, track width, and transition spacing of the data tracks as they are recorded on the media surface. Simply put, linear tape allows greater physical space on the tape for each data bit that is recorded. This reduces the risk of reading/writing errors and data loss. It also means greater opportunities to increase storage capacity in future generations by further increasing areal density and/or decreasing track width and transition spacing.
The use of proven materials such as cobalt-enhanced iron oxides and MP1 tapes also provides linear formats with long potential migration paths through use of advanced materials in future generations.
Performance is most often comparable for linear and helical systems at given capacity points. For example, both Travan NS20 and DDS3 provide 2MBps data transfer rates with 20GB and 24GB, respectively, of compressed capacity per cartridge.
In addition, linear systems make effective use of multi-channel thin-film magneto-resistive heads, which enables further transfer rate improvements by adding channels to the head. For example, Travan NS20 uses a single-channel head, while other systems such as the IBM 3590 have shown the ability to use heads with up to 16 channels. Other technologies such as Overland Data`s VR2 data encoding provide opportunities for dramatic improvements in capacity and performance for linear systems with minimal changes to the drives or media.
In contrast, the relative complexity of helical scan systems and their aggressive use of new tape formulations makes further gains in capacity and performance more difficult and costly to achieve.
Affordability is always somewhat relative, depending on factors such as performance requirements, the mission-critical nature of the data involved, scalability, and so on. But here too, linear-based systems enjoy an inherent advantage–less complex drive mechanisms with fewer moving parts, motors, tape guides, and no tape threading mechanism often translate into a significant cost advantage for linear tape systems as opposed to helical systems at any given capacity point.
It is not surprising then that the market is seeing a proliferation of linear-based tape systems throughout the midrange and enterprise storage markets. From low-cost Travan NS, to midrange SLR and DLT, to future technologies like Ultrium and Accelis, the marketplace is clearly swinging in the direction of linear formats.