Tape backups are an essential and often overlooked part of system administration, regardless of the operating system. Unfortunately, the very nature of tape backup hardware and software has changed so much over the last few years that existing backups routines are often difficult to use properly. Older QIC (quarter inch cartridge) and DC-2000 backup units that held a quarter of a gigabyte of data were great when you only had a few hundred megabytes to back up, but with today’s large capacity drives tape swapping makes the older drives almost unusable for routine backups. Even higher capacity backup systems like DAT (digital audio tape) and 8mm tapes offered only a few gigabytes, often not enough for automated backups for today’s large capacity drives.

Not a lot has happened to tape backup systems to keep up with the increasing capacities of disk drives. DAT has kept up a little, with units like Hewlett-Packard’s SureStore 24e DAT drive offering 24GB of storage. New devices like Sony’s Advanced Intelligent Tape system offer 25GB native and 50GB compressed storage, but these units are expensive. Add to that a high per-tape cost for the AIT drives, and it becomes a pricey alternative for most servers. Fortunately, there is hope. Travan tape backup systems have become more available with 10GB native and 20GB compressed storage, all on a reasonably priced cartridge tape. (The 20GB compressed storage is an average: actual storage depends on the nature of the data.) Add to that a reasonable tape drive cost, and support for almost every UNIX version through utilities like tar and third-party tools like Cheyenne’s ARCServe, and Travan looks even more attractive. Travan units are almost always SCSI-based, so they fit in well with many UNIX systems. (Travan tape drives are also called QIC-3220 tape drives, just in case you see this nomenclature.)

Since Travan tape units are quickly becoming standard on many new SCO UNIX servers, we decided to look at several of the models available on the market. Four vendors (Aiwa, Hewlett-Packard, Seagate and Tecmar) responded to our request for review units. All four have SCO-Certified Travan tape drives, and we mixed two internal and two external units to provide a representative test. Other than packaging and pricing, we were curious as to whether there is a difference in tape drive performance, features, add-in software, and support for other operating systems. While testing Travan units, we also compared their performance to both DAT and AIT backup systems.

Our review platforms were three ALR Revolution 2XL servers each with dual 266MHz Pentium IIs), all three with two 9.1GB Seagate Fast-SCSI drives managed by a DPT SmartRAID IV caching SCSI controller. Each server had 128MB RAM and a Toshiba CD-ROM drive on the same chain as the hard drives. One server was configured with SCO OpenServer 5, another with SCO UnixWare 2.1, and the third with Microsoft’s Windows NT Server 4.0. Using a daughterboard for the DPT SmartRAID IV controllers, we tested each Travan drive on all three servers, once on the same SCSI chain as the hard drives and CD-ROM drive, the second time on a separate SCSI chain. We did this test to see if backup speeds were altered by providing separate chains for the hard drives and tape drive. We also tested the three servers in idle mode (no regular users or background processes) and under normal loaded conditions (ten users with average system loads) to examine backup time changes.

For software on the two SCO boxes we relied on the built-in backup manager (cpio or tar) as well as Cheyenne’s ARCServe Open. Under Windows NT Server we used the Windows NT backup routine and ARCServe IT. For each server, we performed not only the two backup cycles described above, but also tried restoring two files known to be on different halves of the Travan tape. This provided restore timing information for each operating system and backup software combination. We repeated one set of tests for each operating system with the drive in uncompressed mode (our original cycle of tests used the default compression scheme on the drive itself to store 20GB of data). We were curious whether there was a speed difference when the drive was used in compressed and uncompressed mode.

All four Travan tape drives use the same cartridge tapes, one included with each drive unit. The tapes all hold 10GB uncompressed (native) and 20GB with on-board softwaree compression. No special software or drivers is necessary for the compression or the drive itself with either SCO operating system or Windows NT. Both internal Travan drives (Aiwa’s TD-20001 and Seagate’s TapeStor 20) require 50-pin flat SCSI connectors. SCSI IDs are set with jumpers on the back plate of each unit. Both internal drives fit in either a 5.25-inch slot or the smaller 3.5-inch slots. The two external units (Hewlett-Packard’s SureStore T20e and Tecmar’s Travan NS20) use different SCSI connectors (the HP uses SCSI-2 D connectors while the Tecmar uses the older SCSI-1 connector). Both external units allow SCSI IDs to be set with rotary wheels on the back panel. We installed the tape drives as SCSI ID 4 in all tests.

Some older tape drives had reliability problems. System administrators will be happy to know most Travan units have an estimated MTBF (Mean Time Between Failures) of 200,000 hours (the Tecmar is rated at 330,000 hours). The expected lifetime for a Travan tape drive is five years of regular use. Data error rates are estimated to be less than one in 10^15 bits. Travan drives can be mounted vertically or horizontally. We tested all four drives in horizontal mode. The drives don’t differ by much, but a quick examination of each of the units and their shipping package is worthwhile.

Aiwa TD-20001

We received an internal unit from Aiwa, packed with a minimum of fuss and no advertising in a plain brown box. Inside, the Travan unit was wrapped in plastic and accompanied by a spiral-bound Product Manual and a loose-leaf two-page Configuration Guide. There’s not much needed to know about Travan drives in general, so this was plenty of information.

The Aiwa documentation mentions nothing about SCO operating systems, only Windows 95 and Windows 98, but that doesn’t stop UNIX for using the drives. As far as OpenServer and UnixWare are concerned, the Travan drive is just another tape backup unit. Configuring the Travan unit for both SCO operating systems requires only the SCSI ID and maximum tape capacity. A jumper on the back of the Aiwa unit allows you to select compression on or off. We tried both ways, needing only to tell the operating system the maximum capacity.

When a tape is pushed into the drive, there is still about an inch of cartridge extending from the front panel. This may have some effect on servers with doors covering the expansion bays: we could not close the cover of our Revolution 2XLs when a tape was loaded. There is no ejection mechanism: a tug on the drive unloads it (hopefully after a proper rewind: wait for the front-panel light to go out).

Hewlett-Packard SureStore T20e

The HP SureStore T20e was the smaller of the two external units, and the only drive in the test to use SCSI-2 interfaces. It was also the only drive to come with a SCSI cable (a nice touch) although it does require a transformer power block in the power line (as do many HP products). The HP package was by far the most impressive, with a inch-inch thick stack of brochures and manuals included in the deal. Closer inspection shows that there is really only one User’s Guide, repeated in five different languages! Still the User’s Guide is the best produced of the four drives we tested. A CD-ROM accompanying the system includes backup software for Windows and Windows NT.

Loading and unloading tapes is the same as the two internal modes: push and pull. The cartridge stick out the same amount as with the two internals. The internal version of the T20 would have the same front-door problem as the Aiwa and Seagate Travans.

Seagate TapeStor NS20

The Seagate TapeStor NS20 was the smaller of the two internal units but not by much. A small Reference Manual steps you through the installation. The package comes with Seagate’s Backup Exec software, designed for Windows operating systems. This provides a better backup software package than the usual Backup routine with Windows, but it doesn’t have the list of features ARCServe IT offers (although ARCServer IT costs as much as the Travan drive itself).

As with the Aiwa Travan unit, insertion and removal of the tape cartridge is manual. As with the Aiwa, an inch of the cartridge sticks out past the face plate, again an issue if your machine has a door on it.

Tecmar Travan NS20

The Tecmar external NS20 was the largest case of the two externals we tested, and had a pair of the older SCSI-1 connectors on the back of the drive. An on-off switch on the rear allows the drive to be powered down when not needed while a small LED on the front shows drive activity and status. A small fan on the side of the unit keeps the drive’s interior cool. Accompanying the Tecmar drive is a stapled Installation and User’s Guide, and a larger UNIX Environment Installation Guide. The latter contains information for many UNIX workstations, as well as both SCO operating systems. There is nothing of real use in these sections, as they step you through the basic "add tape drive" routines.

One of the first features you notice with the Tecmar drive is what they call iNSync, which is a motorized loading and ejection mechanism. The other drives require physically pushing the cartridge in all the way. The loading of the tape cartridge is analogous to an auto-load VHS deck and is just plain neat. The Tecmar drive also draws the entire tape cartridge case into the drive unit, and slowly ejects it when finished after either the eject button or a software eject command is used.

Testing Results

The four tape drive units and their features are shown in Table 1.

The first issue to deal with when working with any tape backup system is the speed at which a backup is performed. The Travan drives in this test were all attached to SCSI chains, so in theory the throughput would be limited only by the SCSI chain’s throughput (10Mbps on SCSI-1 and 20Mbps on SCSI-2). Sustained speeds will be much slower, of course, because hard drives must be read, data transferred from device to device, and tape writing speeds must be maintained, all playing into the equation. The four manufacturers all estimate an average throughput of 120MB/min when writing compressed mode, and 60MB/min uncompressed (2Mbps and 1Mbps respectively).

The test results for our write operations are shown in Table 2. The Full Backup times are normalized to show any differences based on a backup time of 2 hours 12 minutes. The maximum variation was two minutes, which is scarcely significant. The results for compressed and uncompressed show, not surprisingly, that the uncompressed backup took twice as long as the compressed backup. Since the limiting factor in the backup time is the speed at which data is written, not at which it can be sent to the Travan unit, this is in line with what you would expect. Therefore, there is an intrinsic advantage to backing up with compression not only because of the doubling of tape capacity but also because of the halving of backup time. Is there a disadvantage to compression? It is unlikely in the case of Travan units since the compression is performed on the drive itself, not on software. When backups are compressed in software residing on the operating system, there is always the chance of being unable to read compressed data if the compression software fails. In this case, through, you might as well use the compression system.

Table 2 also shows the comparison between an idle system and a regularly-loaded system (ten users performing normal daily tasks). Since the load imposed by the backup process is minimal we noticed no real difference in backup times between loaded and idle. Keep in mind that we were using dual processors and caching SCSI controllers on our servers, and the effect on a much slower CPU may be more appreciable. To test this assumption we loaded a Pentium Pro 150MHz, 64MB RAM, and a 4GB SCSI hard drive controlled by a standard Adaptec 2940 on-board SCSI controller with SCO OpenServer V and repeated the same test. Ten users performed routine daily tasks over a network to this server. The backups times on the Pentium Pro system were within four minutes of the much faster servers, showing that the backup process is seldom interrupted by the ten user load.

We restored 14MB files from early and late on the tape (the backup log showed the order of files backing up). The restore times were within a few seconds of each other on all four units in both tests, indicating no real performance difference. As expected because of the slow nature of backup units, the times for backups and restores, as well as throughput performance did not change when we moved from a shared SCSI chain to a dedicated chain.

The backup times did not change when we moved from native operating system backup utilities to Cheyenne’s ARCServe Open and ARCServe IT. The latter tools have many more features than the normal backup routines packaged with each operating system, but do nothing to affect performance. However, the ability to back up clients across networks, handle open files properly, and maintain databases of files and tapes backed up is attractive and lends to the popularity of the Cheyenne software. The ability to schedule backups of different types through ARCServe makes it handy, although UNIX users can approximate the same tasks with cron. Still, having used all the alternatives, we would opt for ARCServe every time.

Apart from throughput speeds, other factors contribute to a system administrator’s happiness with tape drives. Speed for rewinding and forwarding to a particular marker are important, too. The Travan units all operated in forward and rewind mode at 120ips. Since the average 10GB Travan tape cartridge has 740 feet of tape, a complete rewind should take 74 seconds. In practice, rewind and fast forward times were close, averaging 83 seconds. Write and read operations on the Travan tapes appear to take place at 100ips at the fastest rate we could measure (we used a calibrated tape with marks every 25 feet for this purpose). Normally, the read and write speeds varied between 44ips and 75 ips.

How does a Travan tape compare to AIT and DAT backups? We timed similar backups from the Aiwa TD-20001 Travan, a Sony SXC-300 AIT, and a Hewlett-Packard SureStore 24 DAT drive. The AIT drive was the fastest of the three, followed by the DAT drive and the Travan drive. The AIT (50GB capacity) took a little less than one hour, the DAT drive (24GB capacity) ninety minutes, and the Travan (20GB Capacity) just over two hours for the same backups. However, the AIT drive is four times as expensive as the Travan, while the DAT is twice as much. You get what you pay for in speed and capacity.

So, all four Travan tape units did the job of backing up and restoring as expected. There is little difference in performance and price (externals are more expensive than internal models), so choosing between the units is a personal choice. The software included with the Seagate and Hewlett-Packard units were impressed for Windows users, but none of the four provides anything special for UNIX. Of the four tape drives, the one that impressed us the most was the Tecmar, primarily because of the nifty loading system. Shallow of us, yeah, but heck, it’s hard to make tape drives sexy. Any one of these four will do you well, regardless of your operating system. Get the best deal you can and base your decision on that.

Table 1: Travan Features

TD-20001 SureStore T20e TapeStor 20 Travan NS20 unit? interface included? Eject button? lights? Table 2: Performance

TD-20001 SureStore T20e TapeStor 20 Travan NS20 Backup (Compressed): System idle minutes minutes minutes minutes Backup (Uncompressed): System idle minutes minutes minutes minutes Backup (Compressed): System busy minutes minutes minutes minutes Backup (Uncompressed): System busy minutes minutes minutes minutes 14MB file from early in tape minutes minutes minutes minutes 14MB file from late in tape minutes minutes minutes minutes