9.2 Choosing a Tape Drive

Consider the following issues when choosing a tape drive:

Capacity

The single most important consideration. Get a drive that can back up all data on one tape, allowing for some growth. If your data set exceeds the capacity of one tape, you may find that drive is no longer usable (if no one is available to change tapes during an overnight backup) or that it has suddenly become very expensive to use (because you must buy twice as many tapes). In such a case, the only alternative to replacing the drive is to use a backup scheme that mixes full and incremental or differential partial backups, which is riskier for your data.

Most tape drive manufacturers arbitrarily rate their drives at double actual native capacity, assuming you will use software or hardware compression to double the amount of data that fits on a tape. AIT drive specifications go further still, assuming 2.6X compression. The actual compression ratio you experience depends on the data mix (e.g., documents and spreadsheets compress well; executables, images, and archives much less so), the backup software you use, and sometimes on the speed of the computer where the drive resides. We find that real-world data sets typically compress at 1.5:1 to 1.7:1, so plan accordingly.

Having adequate tape drive capacity does not necessarily mean being able to back up your entire hard drive (or drive farm) to one tape in one pass. For example, our network contains more than a terabyte (1000 GB) of disk space with hundreds of gigabytes in use. Yet for backups we use only two DDS-3 drives that store 12 GB natively, one DDS-4 drive that stores 20 GB natively and one Travan NS-20 drive that stores 10 GB natively. We are well protected, despite our total tape drive capacity being much smaller than the amount of data we have stored. Why? Because much of that data doesn't need to be backed up frequently, if at all. Some is archived "real" data—things such as previous editions of this book. We back up that data periodically to inexpensive DDS-3 tapes as well as CD-R and DVD+R/RW discs. We need to renew those backups only every few months when we move material from our working data directories to our archive directories. Hundreds of gigabytes of data need not be backed up at all. That comprises copies of operating systems and applications that we keep on network volumes for convenience, ISOs of numerous older Linux distributions, MP3s we ripped from our CD collection (which we plan to re-rip as OGG files anyway), movies we recorded and watched but haven't yet gotten around to deleting, and so on. In other words, stuff that we wouldn't miss if we lost it, or that we could easily reproduce if necessary. Our actual working data fits easily on one DDS-3 tape, so for us DDS-3 suffices.

Speed

This may or may not be a critical factor, depending on your own environment and practices. If you have a limited backup window available, speed may be as important as capacity. If you can simply start a backup when you finish work for the day and allow it to run overnight, speed may be a minor factor.

Actual throughput depends on the drive mechanism, the interface, the speed of the computer in which the drive is installed, and the data set being backed up, but will likely be lower than the drive manufacturer advertises. Compression may also have a significant impact on throughput, for better or worse. For example, our Seagate Scorpion 40 DDS-4 drive is rated at 165 MB/min native and 330 MB/min compressed, but we typically get only 275 MB/min with compression enabled. Similarly, our Seagate Travan NS20 SCSI drive is rated at 60 MB/min native and 120 MB/min compressed, but we actually get about 100 MB/min. The lower-than-expected throughput with compression is no fault of the drives, but results from our data being less compressible than the assumed 2:1. For these drives, compression improves throughput, albeit not to the expected extent. Conversely, when we tested an OnStream DI30 drive that is rated at 60 MB/min native and 120 MB/min with compression, we actually got 45-50 MB/min native, but only 15-17 MB/min when using the bundled Echo software with compression enabled. In that case, it turned out the compression software was poorly designed, a problem that was fixed in a later release. All of these figures are for backing up local volumes. Backing up data across a network, even a fast network, commonly cuts throughput by half or more due to operating system overhead, filesystem overhead, and network latency.

If your tape drive provides hardware compression and your backup application has software compression, don't use both methods together. Either works well by itself, but using both may actually reduce throughput, as the drive tries to compress data that's already been compressed by the backup software. In general, hardware compression is faster than software compression. There are exceptions, though, and which compression method is more efficient may depend on the drive, backup software, and mix of data being backed up. The only way to determine which method is faster is to try both, but be sure to try them one at a time.

Media cost

Travan and OnStream ADR drives are constructed with loose tolerances, and are accordingly inexpensive, but require expensive tapes built to close tolerances. DDS drives, conversely, are expensive because they are built to tight tolerances, which allows them to use inexpensive, loose-tolerance tapes. AIT drives are expensive and use expensive tapes, but their very high capacity and throughput may make them the only alternative when DDS-4 isn't large enough or fast enough. A typical tape rotation may require from four to 50 or more tapes. Tapes must be replaced periodically (on the schedule recommended by the drive and/or tape manufacturer—trying to stretch the lifetime of tapes is a foolish economy). Tape drives have a realistic service life of perhaps two to three years with heavy use, and four or five years with moderate use (by which time the drive is likely no longer adequate for your needs anyway). Expect to spend from as much to several times as much as the cost of the drive to buy tapes over the drive's life.

Interface

Internal tape drives are available with ATAPI/IDE or SCSI interfaces. External tape drives are available with SCSI, Parallel, USB 1.1, USB 2.0, or FireWire (IEEE-1394) interfaces.

ATAPI

ADR and Travan drives are available with ATAPI interfaces. ATAPI drives are typically less expensive than those using other interfaces, can use the ubiquitous IDE interface present on any modern motherboard, provide reasonably high throughput, and are easy to install. Choose an ATAPI drive for convenience, ease of installation, or when cost is an overriding issue.

SCSI

ADR, Travan, DDS, and AIT drives are available with SCSI interfaces. SCSI drives typically sell for at least a $50 to $100 premium over similar ATAPI models, require adding a $100+ SCSI interface card if the PC is not already so equipped, and are more complicated to install and configure than ATAPI models. SCSI drives typically provide much higher throughput and much lower CPU utilization than ATAPI models. The largest and most feature-laden drives are available only in SCSI. Most SCSI-only models are designed for use on servers, and are therefore better built and more reliable than ATAPI drives designed for the mass market. Choose a SCSI drive for highest capacity, performance, durability, and reliability. SCSI is the only option if there are no available ATAPI connections, or if you require capacity and/or features available only in a SCSI model.

Parallel

Drives that use the parallel interface typically have half or less the throughput of SCSI or ATAPI/IDE, but may be a reasonable choice if you must use one drive to back up local data on multiple standalone PCs that do not have recent USB ports. On a small network, it is usually better to map a drive on the server where a tape drive resides for each local workstation volume, and back up centrally to an ATAPI or SCSI drive. If you are considering a parallel drive because you have several standalone PCs that must be backed up, consider instead connecting those PCs with a simple network and using an internal server-based tape drive.

USB

USB 1.1 drives typically have between half and two-thirds the throughput of ATAPI/IDE and SCSI drives, and are a better choice than parallel port drives for backing up local data on multiple standalone PCs if all of those PCs have reasonably recent USB ports. USB 1.1 tape drives have a theoretical maximum throughput of about 90 MB/min and a typical actual throughput of 60 MB/min or less. USB 2.0 models are limited by the speed of the drive mechanism rather than the speed of the interface.

FireWire

There are a few external tape drives available that use the FireWire interface, none of which we have tested. FireWire tape drives offer Plug and Play compatibility similar to USB models, but are much faster than USB 1.1 drives. The relatively high cost of FireWire tape drives and the fact that few PCs have FireWire interfaces make them a poor choice for most people. The exception is if you have desktop PCs and, particularly, notebook PCs that have a FireWire interface installed. In that case, a FireWire tape drive may be the fastest, easiest tool for backing up.

Cross-drive compatibility

Here's a dirty little secret that drive manufacturers don't talk much about. You might reasonably assume that a tape you created in one drive would be readable in a similar drive, but that's not always the case. In particular, we have found that some Travan TR-4 drives produce tapes that cannot be read by another drive, even one of the identical make and model. We seldom encounter that problem on AIT, DDS, Travan NS8/NS20, and OnStream ADR drives, although we have had infrequent reports of such compatibility problems occurring with them.

Unless you use a tape drive to transfer large quantities of data between computers, this may seem a minor issue. It can be critical, however, if your computer is stolen or damaged by flood or fire. Even if your backup tapes are safely locked away, you may find that a replacement drive of the same model is unable to read them. If your data is important enough to warrant extreme precautions, buy two identical tape drives and verify that a tape written in either drive is readable by the other. Repeat this verification periodically because drives do start marching to their own drummer as they age. Alternatively, consider backing up key data frequently to CD-R or writable DVD and verifying that the disc is readable.

Operating system compatibility

Hardware-level compatibility and application compatibility are important issues in choosing a tape drive. At the hardware level, any current tape drive is almost certainly compatible with all recent versions of Windows. Most current tape drives also support Linux, but there are exceptions, so check hardware compatibility and driver availability for your specific Linux distribution and version before you buy a tape drive. Linux drivers may be available for a particular tape drive, but using them may require recompiling the kernel or taking other steps that you may not be comfortable performing.

Application compatibility varies. Many tape drives are bundled with backup software that is targeted at a specific market. For example, a drive intended for desktop systems may include backup software that runs under Windows 9X, 2000 Professional, and XP Professional but not under Windows 2000 Server or Linux. Conversely, a tape drive targeted at servers may include bundled backup software that runs under one or several server operating systems. If you intend to use the tape drive with Linux or an unsupported Windows version, you must obtain backup software separately. That software may be free, as are several Linux backup programs, or it may be expensive enough to make a different tape drive a better choice.